No entries match these filters. Clear filters
CrewAI v1.15.23 adds Gemini 3.8 Flash support and eval tracing fixes
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.23 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.23
CrewAI v1.15.22 adds LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. overlay routing, OpenRouter OpenRouter A unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. embeddings embeddings Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. , and platform catalog access.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.22 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.22
Adds llm_overlay context variable to route agent roles to specific models Not checked yet Adds OpenRouter as a supported embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. provider Not checked yet
2 details in Everything
CrewAI v1.15.21 adds telemetry for checkpoint runtime and CLI usage tracking.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.21 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.21
telemetry captures checkpoint runtime metrics and CLI command usage data. Not checked yet
1 detail in Everything
CrewAI 1.15.19 adds Clipper integration, CEL CEL Common Expression Language, an open-source expression language maintained by Google, used to embed lightweight, safe policy evaluation logic into tools without running a full scripting engine. now() , injectable platform tool client, and machine size reporting.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.19 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.19
Adds Clipper integrations client for connecting to external systems Not checked yet Adds now() function to the CEL expression environment for time-based logic Not checked yet Adds an injectable client for CrewAI platform tools to enable custom service wiring Not checked yet
3 details in Everything
CrewAI 1.15.18 stabilizes conversational flows with new deployment recording and declarative state/router capabilities.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.18 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.18
crewai deploy records the created deployment by its UUID UUID A universally unique identifier, usually a 128-bit value used to identify an object across systems. It is designed to make independently generated identifiers unlikely to collide. Not checked yet Declarative chat flows can now declare their own state shape via state_shape Not checked yet Declarative chat flows can name the router's response format via response_format Not checked yet Conversational declarations accept crew-style LLM config in addition to standard formats Not checked yet project_id is always emitted by project creation commands, distinguishing absent from empty Not checked yet
5 details in Everything
Declarative conversational flows now drive agent interactions with declarations configuration.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.17 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.17
Enable declarations to drive conversational mode in your crew configuration Not checked yet Synthesize built-in conversational methods for declarations via the new config schema Not checked yet
2 details in Everything
CrewAI 1.15.16 adds execution context management with UUID UUID A universally unique identifier, usually a 128-bit value used to identify an object across systems. It is designed to make independently generated identifiers unlikely to collide. support, flow exception recording, and enhanced AMP deployment telemetry.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.16 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.16
Introduces execution context management with UUID support for tracking individual flow and agent executions. Not checked yet Records the type of exception that ended a flow, giving observability pipelines structured failure context. Not checked yet Records when a trace batch is shared with AMP, enabling auditable telemetry handoff visibility. Not checked yet Counts deployments from any origin and records where they started, supporting multi-origin deployment tracking in AMP. Not checked yet
4 details in Everything
CrewAI 1.15.15 adds flow outcome, duration, and human-in-the-loop human-in-the-loop Human in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. signal reporting to flow telemetry.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.15 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.15
CrewAI 1.15.12 adds URLReadTool, unified crewai create <resource> scaffolding, and platform action tool metadata.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.12 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.12
└──▷ TRY IT
Scaffold a new CrewAI resource (crew, flow, etc.) using the unified create command.
copy
$ crewai create <resource>
Unifies project scaffolding under crewai create <resource>, replacing separate scaffold commands. Not checked yet Adds URLReadTool for reading arbitrary URLs inside agent workflows. Not checked yet Adds app metadata to platform action tools. Not checked yet
3 details in Everything
CrewAI 1.15.11 adds an IBM Db2 search tool, enterprise project_id linking, AMP surfacing in AGENTS.md, and coding-agent telemetry detection.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.11 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.11
Adds IBM Db2 search tool, enabling agents to query Db2 databases as a built-in tool. Not checked yet Adds project_id support to link open-source CrewAI usage to an enterprise account. Not checked yet Surfaces CrewAI AMP in AGENTS.md and adds detection of coding agents in telemetry. Not checked yet Tracks interception-hook dispatches in telemetry for improved observability of agent execution. Not checked yet
4 details in Everything
CrewAI 1.15.10 adds skill usage event collection for agent observability.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.10 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.10
Collects skill usage events, enabling observability into which agent skills are invoked during crew runs. Not checked yet
1 detail in Everything
CrewAI 1.15.9 surfaces tool failures explicitly, emits FlowFailedEvent on flow errors, and adds progressive disclosure for skills.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.9 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.9
Surfaces tool failures as actual failures instead of silently reporting them as success, enabling agents to react correctly to tool errors. Not checked yet Emits a new FlowFailedEvent when a flow execution fails, allowing event-driven error handling and observability hooks in Flows. Not checked yet Adds progressive disclosure for skills, surfacing contextually relevant skill guidance incrementally rather than all at once. Not checked yet
3 details in Everything
CrewAI 1.15.8 adds WaitTool to let agents pause execution while waiting on long-running async jobs.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.8 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.8
CrewAI 1.15.7 adds runtime skill usage event emission for observability.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.7 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.7
Emits skill usage events at runtime, enabling observability pipelines to trace which skills agents invoke during execution. Not checked yet
1 detail in Everything
CrewAI 1.15.7a1 adds runtime skill usage event emission for improved observability.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.7a1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.7a1
Emits skill usage events at runtime, enabling observability pipelines to track which skills agents invoke during execution. Not checked yet
1 detail in Everything
CrewAI 1.15.5 adds authentication support for skill registry downloads.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.5 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.5
Supports authenticated downloads from the skill registry, enabling access to private or restricted skills. Not checked yet
1 detail in Everything
CrewAI 1.15.4 promotes the Skills Repository to stable, making AI coding-agent guidance an officially supported feature.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.4 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.4
└──▷ TRY IT
Install the now-stable CrewAI Skills into Claude Code so it automatically applies CrewAI best practices when you scaffold Flows, configure Crews, or design agents.
copy
$ /plugin marketplace add crewAIInc/skills
/plugin install crewai-skills@crewai-plugins
/reload-plugins
Install the stable CrewAI Skills into Cursor, Codex, Windsurf, or any other coding agent via the skills.sh registry.
copy
$ npx skills add crewaiinc/skills
CrewAI 1.15.3 adds step interception hooks with @on, org ID support in PlusAPI, and declarative Flow execution in the TUI TUI A text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. .
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.3 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.3
└──▷ USE IT
Intercept execution boundaries in a crew or flow to log, audit, or gate agent activity at specific lifecycle points.
python
copy
from crewai import on
@on("before_llm_call")
def audit_llm_input(context):
print("LLM about to be called with:", context)
@on("after_llm_call")
def log_llm_output(context):
print("LLM responded with:", context)
Adds step interception points and a generic hook dispatcher, enabling fine-grained execution-boundary callbacks via @on decorators. Not checked yet Adds organization_id parameter to the PlusAPI client for multi-org deployments. Not checked yet Supports running declarative Flows on the TUI with headless terminal fallback. Not checked yet
3 details in Everything
CrewAI 1.15.3a1 adds step interception hooks via @on, declarative flow support in the TUI TUI A text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. , and an organization ID parameter for the PlusAPI client.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.3a1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.3a1
Adds step interception points and a generic hook dispatcher, enabling @on-decorated callbacks at execution boundaries for fine-grained observability and control. Not checked yet Supports running declarative flows on the TUI with a headless terminal fallback. Not checked yet Adds an organization ID parameter to the PlusAPI client for multi-org deployments. Not checked yet Improves custom OpenAI URL handling for self-hosted or proxy LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. endpoints. Not checked yet
4 details in Everything
CrewAI 1.15.2 adds inline skill definitions, templated Flow action inputs, stream frame protocol, and AgentExecutor feedback handling.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.2 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.2
Supports inline skill definitions, eliminating the need for separate skill files in crew/flow configurations. Not checked yet Adds templated Flow action inputs, enabling dynamic parameterization of Flow actions at runtime. Not checked yet Defines a stream frame protocol for Flows, enabling structured streaming output from Flow executions. Not checked yet Adds a generated Flow Definition authoring skill to guide AI coding agents when scaffolding Flows. Not checked yet Implements message setup and feedback handling in AgentExecutor for richer agent interaction loops. Not checked yet
+4 more
show less
Adds repository agents to Flow definitions, enabling agent-backed steps directly in Flow authoring. Not checked yet Adds text helpers for Flow CEL CEL Common Expression Language, an open-source expression language maintained by Google, used to embed lightweight, safe policy evaluation logic into tools without running a full scripting engine. prompts and skill examples, streamlining expression authoring in Flows. Not checked yet Types tool and app fields in CrewDefinition, providing stronger schema validation for crew configs. Not checked yet Pulls latest LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. models dynamically in the crew wizard, keeping model selection current without manual updates. Not checked yet
9 details in Everything
CrewAI 1.15.2a1 adds inline skill definitions, a stream frame protocol for Flows, and new tool/app types in CrewDefinition.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.2a1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.2a1
Supports inline skill definitions, enabling skills to be declared directly within agent or crew configurations without separate files. Not checked yet Introduces a stream frame protocol for Flows, defining a structured message format for streaming Flow execution events. Not checked yet Adds tool and app type options to CrewDefinition, expanding the kinds of crew artifacts that can be declared. Not checked yet Adds a generated Flow Definition authoring skill for AI coding agents, guiding them when scaffolding or editing Flow definitions. Not checked yet
4 details in Everything
CrewAI 1.15.1 auto-initializes Git repos for new projects, enforces explicit crew definitions, and opens the deployment page post-deploy.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.1
Automatically initializes a Git repository when generating a new CrewAI project via the CLI. Not checked yet Requires explicit CrewAI project definitions, enforcing structured crew configuration at project creation. Not checked yet Opens the deployment page automatically in the browser after a successful crewai deploy command. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! Explicit CrewAI project definitions are now required (#6358)
CrewAI 1.15.1a1 adds TUI TUI A text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. button telemetry, enforces explicit project definitions, and auto-opens the deployment page after CLI deploy.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.1a1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.1a1
Adds telemetry tracking for TUI button interactions, enabling usage analytics on crew terminal UI actions. Not checked yet Requires explicit CrewAI project definitions, enforcing stricter project structure as a breaking change for existing setups. Not checked yet Automatically opens the deployment page in the browser after a successful crewai deploy CLI command. Not checked yet
3 details in Everything
CrewAI 1.15.0 adds declarative Flow definitions with crew/agent actions, each composite steps, conditional expressions, and DMN mode support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.15.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.15.0
Adds unified declarative Flow loading, letting flows be defined and loaded from configuration rather than pure Python code. Not checked yet Adds declarative Flow CLI support so flows defined in config files can be kicked off directly from the command line. Not checked yet Adds each composite action to FlowDefinition for iterating over collections within a flow step. Not checked yet Adds optional if expression to each.do steps, enabling conditional execution inside iteration blocks. Not checked yet Adds single-agent action support in Flow definitions, allowing a lone agent call to be wired directly into a flow without a full crew. Not checked yet
+7 more
show less
Adds crew actions to FlowDefinition, enabling inline crew execution as a named step in a declarative flow. Not checked yet Adds inline crew definition loading so a crew can be fully specified inside a FlowDefinition without a separate crew config file. Not checked yet Implements DMN (Decision Model and Notation) mode support in crew creation and execution for structured decision logic. Not checked yet Tracks conversational flow turn usage in telemetry for visibility into multi-turn conversation costs. Not checked yet Supports conversational flows in the CLI TUI TUI A text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. , surfacing turn-by-turn flow progress in the terminal UI. Not checked yet Consolidates crewai run and crewai flow kickoff into a single command entry point (breaking change for scripts that invoked them separately). Not checked yet Removes StateProxy from flow state access, providing direct state access in flow methods (breaking change). Not checked yet
12 details in Everything
└──▷ BREAKING ON UPGRADE
! The StateProxy has been removed from flow state access
CrewAI 1.14.8a5 enables declarative refs to work seamlessly across both Flows and Crews.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.8a5 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.8a5
Enables declarative references to be resolved across Flow and Crew boundaries, allowing shared state and configuration to be referenced from either context. Not checked yet
1 detail in Everything
CrewAI 1.14.8a4 adds conversational flow support in the CLI TUI TUI A text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. .
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.8a4 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.8a4
Supports conversational flows in the CLI text-based UI (TUI), enabling interactive multi-turn flow execution from the terminal. Not checked yet
1 detail in Everything
CrewAI 1.14.8a3 adds declarative Flow loading, unified CLI commands, router-as-start support, and typed tool output schemas.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.8a3 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.8a3
└──▷ USE IT
Define a Flow whose entry point is a router for immediate conditional branching at startup.
python
copy
from crewai.flow.flow import Flow, router, listen, start
class TriageFlow(Flow):
@router()
@start()
def triage(self):
if self.state.priority == 'high':
return 'urgent'
return 'normal'
@listen('urgent')
def handle_urgent(self): ...
@listen('normal')
def handle_normal(self): ...
Adds unified declarative Flow loading, enabling Flows to be defined and loaded via configuration rather than only imperative Python. Not checked yet Consolidates crewai run and crewai flow kickoff into a single command, simplifying the CLI surface. Not checked yet Adds declarative Flow CLI support, letting practitioners launch declaratively defined Flows from the command line. Not checked yet Enables @router() to be used as the start method of a Flow, allowing conditional branching logic at Flow entry points. Not checked yet Adds typed output schemas for CrewAI tools, enabling structured, validated outputs from tool calls. Not checked yet
+1 more
show less
Keeps flow method progress visible for nested Crews, improving observability when Crews are embedded inside Flows. Not checked yet
6 details in Everything
CrewAI 1.14.8a2 adds single-agent actions inside Flow definitions and CEL CEL Common Expression Language, an open-source expression language maintained by Google, used to embed lightweight, safe policy evaluation logic into tools without running a full scripting engine. expression validation at load time.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.8a2 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.8a2
Adds support for single agent actions directly within Flow definitions, enabling lightweight LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. calls alongside full Crew steps without spinning up a multi-agent Crew. Not checked yet Validates Flow CEL (Common Expression Language) expressions at definition load time, catching misconfigured routing conditions before runtime. Not checked yet
2 details in Everything
CrewAI 1.14.8a1 adds optional if expressions to each.do loop steps in Flows.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.8a1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.8a1
Supports optional if conditional expressions on each.do steps, enabling per-iteration filtering without wrapping logic in separate tasks. Not checked yet
1 detail in Everything
CrewAI 1.14.8a introduces JSON-first crew definitions, FlowDefinition actions, DMN mode, and a new crewai run --definition CLI flag.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.8a https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.8a
└──▷ TRY IT
Run a flow defined entirely in a JSON/definition file — no Python code required — to execute a crew-based automation.
copy
$ crewai run --definition
Adds crewai run --definition experimental CLI flag to run flows directly from a flow definition file without writing Python code. Not checked yet Introduces JSON-first crew projects, enabling crew creation, configuration, and execution entirely through JSON rather than Python. Not checked yet Adds script/code block, crew, and each composite actions to FlowDefinition, enabling richer declarative flow composition. Not checked yet Adds expression support to FlowDefinition actions, enabling dynamic values within declarative flow definitions. Not checked yet Implements DMN (Decision Model and Notation) mode support in crew creation and execution. Not checked yet
+4 more
show less
Enables human feedback to be driven directly from a flow definition, removing the need for custom Python callbacks. Not checked yet Wires config and persistence settings from FlowDefinition into the runtime, closing the gap between declarative definitions and live execution. Not checked yet Supports ZIP deployment fallback and JSON crew project environment runs, enabling portable crew packaging and deployment. Not checked yet Enhances memory reset functionality with improved JSON crew handling. Not checked yet
9 details in Everything
CrewAI 1.14.7 adds a chat API, Snowflake Cortex LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. , pluggable memory/RAG RAG Retrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. backends, richer LLM event metadata, and route-aware Flow decorators.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.7 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.7
└──▷ USE IT
Route a flow's next step based on a condition using the new route-aware DSL decorator syntax.
python
copy
from crewai.flow.flow import Flow, listen, router, start
class TriageFlow(Flow):
@start()
def ingest(self):
self.state.severity = classify_alert()
@router(ingest)
def route_by_severity(self):
if self.state.severity == "critical":
return "escalate"
return "log_only"
@listen("escalate")
def escalate(self):
...
@listen("log_only")
def log_only(self):
...
Use Snowflake Cortex as the LLM provider for an agent that queries internal data warehouses.
python
copy
from crewai import Agent, LLM
analyst = Agent(
role="Data Analyst",
goal="Summarize query results from Snowflake",
backstory="Expert in SQL and data warehousing.",
llm=LLM(model="snowflake/snowflake-arctic-instruct"),
)
Adds a chat API for building conversational flows, enabling back-and-forth interaction patterns within CrewAI Flows. Not checked yet Adds native Snowflake Cortex as a supported LLM provider. Not checked yet Introduces pluggable default backends for memory, knowledge, RAG, and flow — allowing custom storage and retrieval engines without subclassing core internals. Not checked yet Surfaces real finish_reason, sampling parameters, and response.id on LLM events for richer observability. Not checked yet Introduces route-aware decorators for typing DSL triggers in Flows, enabling cleaner conditional routing definitions. Not checked yet
+4 more
show less
Supports building a FlowDefinition from Flow DSL metadata, exposing flow structure programmatically. Not checked yet Makes the locking backend overridable for concurrent-run isolation and custom synchronization strategies. Not checked yet Adds crew trained agents file support for persisting and reloading agent training state. Not checked yet Improves import speed via lazy-loading of docling — meaningfully reducing cold-start latency in large deployments. Not checked yet
9 details in Everything
CrewAI 1.14.6 adds env-var isolation in StdioTransport, Databricks env_var declarations, and enhanced planning configuration.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.6 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.6
Adds env_vars declaration support on DatabricksQueryTool for explicit credential scoping. Not checked yet Improves planning configuration and observation handling in agent execution. Not checked yet Moves Skills Repository behind the CREWAI_EXPERIMENTAL gate as an experimental feature. Not checked yet
3 details in Everything
CrewAI 1.14.5 adds state-restore kickoff, ExaSearchTool highlights, and Daytona sandbox improvements.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.5 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.5
└──▷ USE IT
Resume a long-running Crew from a checkpoint state ID instead of restarting from scratch.
python
copy
result = my_crew.kickoff(restore_from_state_id="<state_id>")
Adds restore_from_state_id kickoff parameter to resume a Crew run from a previously saved state. Not checked yet Adds highlight results support to ExaSearchTool (formerly EXASearchTool). Not checked yet Improves Daytona sandbox tool integration. Not checked yet Defaults Crew agents to AgentExecutor, deprecating CrewAgentExecutor. Not checked yet
4 details in Everything
└──▷ BREAKING ON UPGRADE
! The EXASearchTool is renamed to ExaSearchTool; imports using the old name will break.
CrewAI 1.14.4 adds custom persist keys, Azure OpenAI Responses API, Vertex AI workload identity workload identity An identity assigned to software, such as a service or container, rather than a person. It lets the workload authenticate and receive permissions without sharing a user's credentials. , Tavily Research, and You.com MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tools.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.4 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.4
Supports custom persistence keys in the @persist decorator for fine-grained checkpoint control. Not checked yet Adds Responses API support for the Azure OpenAI provider. Not checked yet Forwards credential_scopes to the Azure AI Inference client for scoped authentication. Not checked yet Adds Vertex AI workload identity setup for credential-free GCP deployments. Not checked yet Adds Tavily Research and Get Research tools for AI-powered web research workflows. Not checked yet
+1 more
show less
Adds You.com MCP tools covering search, research, and content extraction. Not checked yet
6 details in Everything
CrewAI 1.14.3 adds lifecycle events for checkpoints, e2b/Daytona sandbox support, Bedrock V4, Azure DefaultAzureCredential fallback, and ~29% cold-start reduction.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.3 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.3
└──▷ USE IT
Use keyless Azure auth in environments where injecting an API key is not possible (e.g., managed identity on Azure).
python
copy
from crewai import LLM
# No api_key provided — CrewAI falls back to DefaultAzureCredential automatically
llm = LLM(
model="azure/<your-deployment>",
azure_endpoint="https://<your-resource>.openai.azure.com/",
api_version="2024-02-01",
)
Adds lifecycle events for checkpoint operations, enabling observability hooks around checkpoint create, resume, and fork actions. Not checked yet Supports e2b sandbox integration for secure code execution within crews. Not checked yet Adds Daytona sandbox tools for enhanced sandboxed functionality. Not checked yet Adds checkpoint and fork support to standalone agents, not just full crews. Not checked yet Falls back to DefaultAzureCredential when no API key is provided in the Azure integration, enabling keyless auth flows. Not checked yet
+2 more
show less
Adds Bedrock V4 support for AWS-hosted model access. Not checked yet Optimizes MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. SDK and event types to reduce cold start by ~29%, unlocking faster agent startup in latency-sensitive deployments. Not checked yet
7 details in Everything
CrewAI 1.14.2 adds checkpoint resume/fork/prune, template management, and enriched LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. token tracking.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.2 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.2
└──▷ USE IT
Resume a failed crew run from a checkpoint instead of restarting from scratch.
python
copy
result = crew.kickoff(from_checkpoint=<checkpoint_id>)
Adds checkpoint resume, diff, and prune commands for managing crew execution checkpoints. Not checked yet Adds from_checkpoint parameter to Agent.kickoff and related methods to restart runs from a saved state. Not checked yet Adds checkpoint forking with lineage tracking, enabling branched execution from any saved checkpoint. Not checked yet Adds template management commands for creating and managing project templates. Not checked yet Adds deploy validation CLI for pre-flight checks before deploying crews. Not checked yet
+1 more
show less
Enriches LLM token tracking with reasoning tokens reasoning tokens Intermediate reasoning a model writes out before its final answer. Reasoning models generate it automatically and it is often billed as extra output tokens. and cache creation tokens. Not checked yet
6 details in Everything
CrewAI 1.14.1 adds async checkpoint TUI TUI A text-based user interface that runs inside a terminal, using characters and ANSI escape codes to render menus, panes, and widgets without a graphical desktop, letting cyber tools ship rich interactivity with no GUI dependency. browser and proper resource management for streaming outputs.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.1
Adds async checkpoint TUI browser for interactive inspection of crew checkpoints. Not checked yet Adds aclose() /close() methods and async context manager support to streaming outputs for proper resource cleanup. Not checked yet
2 details in Everything
CrewAI 1.14.0 adds runtime state checkpointing with SQLite storage, automatic checkpoint config, and new CLI commands for checkpoint management.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.14.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.14.0
└──▷ TRY IT
List all saved checkpoints for a crew run to find a resume point after an interrupted execution.
copy
$ crewai checkpoint list
Inspect details of a specific checkpoint before deciding whether to resume from it.
copy
$ crewai checkpoint info <checkpoint-id>
Adds checkpoint list and checkpoint info CLI commands to inspect saved crew execution states. Not checked yet Introduces SqliteProvider for persistent checkpoint storage backend. Not checked yet Adds CheckpointConfig for configuring automatic checkpointing of crew runs. Not checked yet Implements runtime state checkpointing and an event system so long-running crews can be resumed after interruption. Not checked yet Adds guardrail_type and name fields to traces to distinguish guardrail events in observability pipelines. Not checked yet
+1 more
show less
Exports JsonProvider as an additional checkpoint storage option alongside SqliteProvider. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
! CodeInterpreterTool is removed and code execution parameters are deprecated — any crew or agent config referencing CodeInterpreterTool or those parameters will break on upgrade.
CrewAI 1.13.0 adds unified runtime state serialization, richer telemetry spans, token usage events, and an A2UI extension.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.13.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.13.0
└──▷ USE IT
Capture token usage from every LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. call — useful for cost attribution or rate-limit monitoring in a pipeline.
python
copy
from crewai.events import LLMCallCompletedEvent
from crewai.event_bus import event_bus
@event_bus.on(LLMCallCompletedEvent)
def on_llm_done(event: LLMCallCompletedEvent):
print(event.token_usage)
Adds RuntimeState RootModel for unified state serialization across crew runs. Not checked yet Enhances the event listener with new telemetry spans covering skill and memory events. Not checked yet Adds A2UI extension with v0.8/v0.9 support, schemas, and documentation. Not checked yet Emits token usage data in LLMCallCompletedEvent for per-call cost and usage tracking. Not checked yet Reduces framework overhead via a lazy event bus and skips tracing entirely when disabled. Not checked yet
5 details in Everything
CrewAI 1.12.1 adds Qdrant memory backend, agent skills, native OpenAI-compatible providers, and automatic hierarchical memory isolation.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.12.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.12.1
└──▷ USE IT
Connect CrewAI to a self-hosted Ollama instance as a native OpenAI-compatible provider, bypassing LiteLLM.
python
copy
from crewai import LLM
llm = LLM(
model="ollama/llama3",
base_url="http://localhost:11434"
)
Adds request_id to HumanFeedbackRequestedEvent for tracking human-in-the-loop human-in-the-loop Human in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. feedback requests. Not checked yet Adds Qdrant Edge storage backend for the memory system. Not checked yet Adds agent skills support. Not checked yet Implements automatic root_scope for hierarchical memory isolation across agents. Not checked yet Supports native OpenAI-compatible providers: OpenRouter OpenRouter A unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. , DeepSeek, Ollama, vLLM, Cerebras, and Dashscope. Not checked yet
+2 more
show less
Adds logout command to the CLI. Not checked yet Adds docs-check command to analyze changes and generate docs with translations. Not checked yet
7 details in Everything
CrewAI 1.12.0 adds Qdrant memory backend, agent skills, native OpenAI-compatible providers, hierarchical memory isolation, and a CLI logout command.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.12.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.12.0
└──▷ TRY IT
Log out of the CrewAI platform from the terminal to rotate credentials or switch accounts.
Adds Qdrant Edge as a storage backend for the memory system. Not checked yet Introduces agent skills as a new first-class capability for agents. Not checked yet Implements automatic root_scope for hierarchical memory isolation across crews. Not checked yet Supports native OpenAI-compatible providers: OpenRouter OpenRouter A unified API gateway that routes requests to dozens of LLM providers under one endpoint, letting cyber tools swap or compare models without rewriting provider-specific integration code. , DeepSeek, Ollama, vLLM, Cerebras, and Dashscope — without requiring LiteLLM. Not checked yet Adds a logout command to the CrewAI CLI. Not checked yet
+1 more
show less
Adds a docs-check CLI command to analyze changes and generate translated documentation. Not checked yet
6 details in Everything
CrewAI 1.11.1 adds flow_structure() serializer for programmatic Flow class introspection.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.11.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.11.1
└──▷ USE IT
Inspect the structure of a defined Flow programmatically — useful for debugging or generating visual representations of agent pipelines.
python
copy
flow = MyFlow()
structure = flow.flow_structure()
print(structure)
CrewAI 1.10.0 adds user input handling in Flows, enhanced MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. tool resolution, and auto-updating tool specs.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.10.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.10.0
Adds user input handling in Flows, enabling human-in-the-loop human-in-the-loop Human in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. data collection mid-flow execution. Not checked yet Enhances MCP tool resolution and emits related events for better observability of MCP-based tool use tool use A model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. . Not checked yet Adds started_event_id tracking on the event bus for correlating flow and agent lifecycle events. Not checked yet Introduces auto-update of tools.specs, keeping tool specifications current without manual intervention. Not checked yet Enhances HITL HITL Human in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. (Human-in-the-Loop) self-loop functionality in human feedback integration. Not checked yet
+4 more
show less
Improves JSON argument parsing and validation in CrewAgentExecutor and BaseTool for more robust tool invocation. Not checked yet Migrates CLI HTTP client from requests to httpx, enabling async-compatible CLI operations. Not checked yet Adds versioned documentation support and yanked-version detection in release notes. Not checked yet Updates LanceDB version and adds lance-namespace packages for expanded vector store vector store A database built to store embeddings and find the nearest ones to a query quickly. It is the retrieval half of most RAG systems. integration. Not checked yet
9 details in Everything
CrewAI 1.9.1 adds before/after tool call hooks and structured output structured output Forcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed. support across providers
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.9.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.9.1
Adds before and after tool call hooks in CrewAgentExecutor for intercepting and acting on tool invocations. Not checked yet Supports structured outputs structured outputs Forcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed. and response_format parameter across LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. providers for typed agent responses. Not checked yet
2 details in Everything
CrewAI 1.9.0 adds A2A A2A Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. task execution, Keycloak SSO SSO Single sign-on, using one identity provider to sign in to multiple applications. It reduces separate passwords and centralizes account and access management. , structured outputs structured outputs Forcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed. , and native multimodal multimodal Able to work with more than one type of input or output, such as text, images and audio. A multimodal model can combine information across those formats. file handling.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.9.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.9.0
└──▷ USE IT
Enforce a typed JSON schema on LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. output so downstream code can safely parse structured results.
python
copy
from pydantic import BaseModel
from crewai import Agent, Task, Crew
class ThreatReport(BaseModel):
severity: str
affected_systems: list[str]
recommendation: str
analyst = Agent(role="Threat Analyst", goal="Analyze security incidents", backstory="...", verbose=False)
task = Task(
description="Analyze the attached incident log and produce a threat report.",
expected_output="A structured threat report",
agent=analyst,
response_format=ThreatReport,
)
crew = Crew(agents=[analyst], tasks=[task])
result = crew.kickoff()
report: ThreatReport = result.pydantic
Adds structured outputs and response_format support across providers for typed, schema-constrained LLM responses. Not checked yet Adds response_id field in streaming responses for tracking and correlating streamed completions. Not checked yet Adds event ordering and parent-child hierarchy to the event system for richer execution tracing. Not checked yet Adds Keycloak SSO provider support for enterprise authentication integration. Not checked yet Adds native multimodal file handling with OpenAI Responses API support. Not checked yet
+5 more
show less
Adds agent-to-agent (A2A) task execution utilities for orchestrating inter-agent workflows. Not checked yet Adds A2A server config and agent card generation for declaring and advertising agent capabilities. Not checked yet Adds additional A2A events with enriched event metadata for deeper observability into A2A interactions. Not checked yet Adds additional A2A transports for flexible agent-to-agent communication. Not checked yet Adds Galileo integration for LLM evaluation and observability. Not checked yet
10 details in Everything
CrewAI 1.8.1 adds Agent-to-Agent Agent-to-Agent Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. (A2A A2A Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. ) task execution, server configuration, agent card generation, and new transports.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.8.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.8.1
Adds A2A task execution utilities for orchestrating agent-to-agent workflows. Not checked yet Adds A2A server configuration and agent card generation to expose CrewAI agents as A2A-compatible services. Not checked yet Adds additional A2A transports, broadening connectivity options between agents. Not checked yet Adds Galileo to the integrations page, enabling Galileo observability for CrewAI workflows. Not checked yet
4 details in Everything
CrewAI 1.8.0 adds Agent-to-Agent Agent-to-Agent Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. async/streaming/push update mechanisms and Human-in-the-Loop support for Flows.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.8.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.8.0
Adds native async chain support for agent-to-agent (A2A A2A Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. ) communication. Not checked yet Introduces A2A update mechanisms — poll, stream, and push — with configurable handlers. Not checked yet Adds Human-in-the-Loop (HITL HITL Human in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. ) feedback support directly within Flows via global flow configuration. Not checked yet Adds streaming tool call events for real-time observability of tool execution. Not checked yet Introduces production-ready Flows and Crews architecture. Not checked yet
+1 more
show less
Improves EventListener and TraceCollectionListener for enhanced event handling. Not checked yet
6 details in Everything
CrewAI 1.7.1 adds a --no-commit flag to the bump command and switches to JSON schema for tool argument serialization.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.7.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.7.1
└──▷ TRY IT
Bump the project version without creating a git commit, useful in CI pipelines where you control commits separately.
copy
$ crewai bump --no-commit
Adds --no-commit flag to the bump command, allowing version bumps without auto-committing changes. Not checked yet Switches tool argument serialization to use JSON schema for more structured, interoperable tool definitions. Not checked yet
2 details in Everything
CrewAI 1.7.0 adds comprehensive async support across flows, crews, tasks, tools, memory, knowledge, and LLMs LLMs A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. .
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.7.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.7.0
└──▷ USE IT
Run a full crew asynchronously to avoid blocking the event loop in an async application or service.
python
copy
import asyncio
from crewai import Crew, Agent, Task
agent = Agent(role="Analyst", goal="Analyze data", backstory="Expert analyst")
task = Task(description="Summarize the report", agent=agent, expected_output="Summary")
crew = Crew(agents=[agent], tasks=[task])
async def main():
result = await crew.kickoff_async()
print(result)
asyncio.run(main())
Kick off a Flow asynchronously so multiple flows can run concurrently in the same process.
python
copy
import asyncio
from crewai.flow.flow import Flow, start
class MyFlow(Flow):
@start()
async def run_step(self):
return "done"
async def main():
flow = MyFlow()
result = await flow.kickoff_async()
print(result)
asyncio.run(main())
Adds async flow kickoff, enabling non-blocking orchestration of multi-agent pipelines. Not checked yet Adds async crew support so entire crews can be run concurrently without blocking the event loop. Not checked yet Adds async task support for individual task execution within crews. Not checked yet Adds async knowledge and memory support for non-blocking retrieval and storage operations. Not checked yet Adds async support for tools and the agent executor with improved typing. Not checked yet
+4 more
show less
Adds native async tool support, allowing tools to be defined and invoked asynchronously. Not checked yet Adds async LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. support for non-blocking model inference calls. Not checked yet Implements the agent-to-agent (A2A A2A Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. ) extensions API with async agent card caching. Not checked yet Introduces system event types and a handler for observability and lifecycle events. Not checked yet
9 details in Everything
CrewAI 1.6.0 adds streaming results for flows and crews, Entra ID Entra ID Microsoft's cloud-based identity and access management service, formerly Azure Active Directory, used by cyber tools to authenticate users and enforce access policies in Microsoft 365 and Azure environments. CLI login, a Merge Agent Handler tool, and Gemini 2.5 Pro preview support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.6.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.6.0
Adds streaming result support to flows and crews, enabling real-time output consumption. Not checked yet Supports CLI login with Entra ID (Microsoft Entra / Azure AD) for enterprise SSO SSO Single sign-on, using one identity provider to sign in to multiple applications. It reduces separate passwords and centralizes account and access management. authentication. Not checked yet Adds Merge Agent Handler tool for combining agent outputs. Not checked yet Enhances flow event state management for improved control over flow lifecycle events. Not checked yet
4 details in Everything
CrewAI 1.5.0 adds LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. call hooks, exposes messages to task outputs, and improves A2A A2A Agent2Agent, an open protocol started by Google for AI agents from different vendors to discover each other, delegate tasks and exchange results. It complements MCP, which connects an agent to tools. trust and Okta data.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.5.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.5.0
└──▷ USE IT
Inspect the full message history attached to a task result for auditing or downstream processing.
python
copy
result = crew.kickoff()
for task_output in result.tasks_output:
print(task_output.messages)
Adds before and after LLM call hooks in CrewAgentExecutor for observability and control around every LLM invocation. Not checked yet Exposes messages to TaskOutput and LiteAgentOutputs, giving callers access to the full message history from a task run. Not checked yet Adds A2A trust remote completion status flag for agent-to-agent trust workflows. Not checked yet Fetches and stores richer data about Okta authorization servers for enhanced auth integrations. Not checked yet Enhances schema description of QdrantVectorSearchTool for improved usability and discoverability. Not checked yet
5 details in Everything
CrewAI 1.4.0 adds first-class MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. support, LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. message interceptor hooks, and enhances the QdrantVectorSearchTool.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.4.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.4.0
Adds first-class support for MCP (Model Context Protocol Model Context Protocol Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. ) with improved concurrent tool execution handling. Not checked yet Enhances QdrantVectorSearchTool with new capabilities for vector search workflows. Not checked yet Adds support for non-AST AST An abstract syntax tree, a structured representation of parsed code. Tools inspect or transform this tree to understand expressions and statements rather than just matching text. plot routes, expanding flow visualization options. Not checked yet Caches i18n prompts for more efficient repeated use across agents. Not checked yet
4 details in Everything
CrewAI 1.3.0 enhances the QdrantVectorSearchTool with improved vector search capabilities.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.3.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.3.0
CrewAI 1.2.1 adds Datadog observability integration and MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. /app support in LiteAgent.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.2.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.2.1
Adds Datadog integration for observability and tracing of CrewAI agent runs. Not checked yet Enables apps and MCPs (Model Context Protocol Model Context Protocol Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers) as supported resources in LiteAgent. Not checked yet
2 details in Everything
CrewAI 1.1.0 adds multi-provider LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. support in InternalInstructor, a mypy plugin base, and improves QdrantVectorSearchTool.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.1.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.1.0
Adds multi-provider LLM support to InternalInstructor, enabling structured output structured output Forcing a model's response to match a given format or JSON schema, so software can parse it reliably instead of hoping the text is well formed. extraction across different LLM backends. Not checked yet Introduces a mypy plugin base for improved static type checking of CrewAI code. Not checked yet Improves QdrantVectorSearchTool for better vector search integration. Not checked yet
3 details in Everything
CrewAI 1.0.0 adds enhanced knowledge/guardrail event handling and tool repository credential injection for the run command.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.0.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 1.0.0
Enhances knowledge and guardrail event handling in the Agent class for more robust agent lifecycle control. Not checked yet Injects tool repository credentials automatically in the crewai run command, enabling authenticated tool source access. Not checked yet
2 details in Everything
CrewAI 0.201.1 adds ChromaDB support for WatsonX and VoyageAI embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. providers.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.201.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.201.1
└──▷ BREAKING ON UPGRADE
! The watson embedding provider is renamed to watsonx; environment variable prefixes are updated accordingly — any config using the old watson provider name or its env var prefixes will break on upgrade.
CrewAI 0.201.0 adds a crewai uv CLI wrapper, custom embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. types, thread-safe platform context, and failed-trace marking for observability.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.201.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.201.0
└──▷ USE IT
Set batch_size on an embedder to avoid hitting token limits when indexing large document sets.
python
copy
embedder_config = {
"provider": "openai",
"config": {
"model": "text-embedding-3-small",
"batch_size": 50
}
}
Adds crewai uv wrapper command to invoke uv directly through the CrewAI CLI. Not checked yet Supports batch_size configuration on embedders to stay within token limits. Not checked yet Introduces thread-safe platform context management for concurrent crew workflows. Not checked yet Enables marking traces as failed in observability workflows. Not checked yet Adds custom embedding types and provider migration support. Not checked yet
+3 more
show less
Upgrades ChromaDB dependency to v1.1.0 with type improvements. Not checked yet Introduces Pydantic v2 support along with pydantic-settings and dependency group reorganization. Not checked yet Adds Pydantic-compatible import validation, replacing legacy utilities. Not checked yet
8 details in Everything
CrewAI 0.193.0 makes RAG RAG Retrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. /knowledge/memory search params fully configurable and enables ChromaDB OpenAI embeddings embeddings Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. .
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.193.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.193.0
Makes search parameters for RAG, knowledge, and memory fully configurable. Not checked yet Enables ChromaDB to use OpenAI API as an embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. function. Not checked yet Adds deeper observability tools for user-level insights. Not checked yet Introduces thread-safe platform context management. Not checked yet Unifies RAG storage system with instance-specific client support. Not checked yet
+2 more
show less
Adds ephemeral trace improvements for better trace control. Not checked yet Adds missing event exports to __init__.py for consistent module behavior. Not checked yet
7 details in Everything
CrewAI 0.186.0 adds Qdrant RAG RAG Retrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. support, partial flow resumability, auto-injection of trigger payloads, and a new config reset command.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.186.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.186.0
└──▷ TRY IT
Reset stored auth/API tokens without being prompted for email — useful when rotating credentials in CI.
copy
$ crewai config reset
Adds support for crewai_trigger_payload auto-injection into crew/flow executions. Not checked yet Introduces a centralized RAG configuration system with optional imports, including Qdrant as a new RAG provider. Not checked yet Implements generic clients for ChromaDB and Qdrant storage backends. Not checked yet Enables partial flow resumability, allowing interrupted flows to resume mid-execution. Not checked yet Displays task names in verbose output for clearer runtime observability. Not checked yet
+3 more
show less
Introduces crewai config reset command for resetting authentication tokens. Not checked yet Adds additional parameters to Flow.start() methods for more flexible flow invocation. Not checked yet Introduces centralized configuration for embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. types and a base embedding client. Not checked yet
8 details in Everything
└──▷ BREAKING ON UPGRADE
! A deprecation warning is now emitted for Task.max_retries; callers using this field should migrate before it is removed.
CrewAI 0.175.0 adds Qdrant RAG RAG Retrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. support, crewai config reset, auto-injected trigger payloads, and a centralized embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. client system.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.175.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.175.0
└──▷ TRY IT
Clear cached login tokens from the local crewAI config when rotating credentials or switching accounts.
copy
$ crewai config reset
Adds Qdrant as a RAG provider, with a new generic Qdrant client alongside an existing ChromaDB client. Not checked yet Introduces a centralized embedding types system and a base embedding client for standardized embedding configuration. Not checked yet Adds crewai config reset command to clear stored tokens. Not checked yet Enables crewai_trigger_payload auto-injection into flows triggered by automation. Not checked yet Adds support for additional parameters in Flow.start() methods. Not checked yet
+3 more
show less
Displays task names in verbose CLI output for better observability. Not checked yet Simplifies RAG client initialization with a new RAG configuration system. Not checked yet Adds support to remove Auth0 and email entry on crewai login. Not checked yet
8 details in Everything
CrewAI 0.165.1 adds agent message history to ExternalMemory, auto-injects trigger payloads, and links memory entries to agent IDs in Mem0.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.165.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.165.1
Includes exchanged agent messages in ExternalMemory metadata for richer memory context. Not checked yet Automatically injects crewai_trigger_payload into crew execution context. Not checked yet Adds support for agent_id-linked memory entries in Mem0 integration. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! The internal flag inject_trigger_input is renamed to allow_crewai_trigger_context; any code or config referencing the old name will break.! The AgentOps integration has been removed; any setup relying on it will no longer function.
CrewAI 0.165.0 enriches ExternalMemory with agent messages, auto-injects trigger payloads, and adds agent_id-linked Mem0 entries.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.165.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.165.0
Includes exchanged agent messages in ExternalMemory metadata for richer memory context. Not checked yet Automatically injects crewai_trigger_payload into crew runs triggered externally. Not checked yet Adds support for agent_id-linked memory entries in Mem0 integration. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! The internal flag inject_trigger_input is renamed to allow_crewai_trigger_context; any configuration or code referencing inject_trigger_input will break.! The AgentOps integration has been removed; any setup relying on it will stop working.
CrewAI 0.159.0 adds enterprise CLI setup command and partial flow resumability support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.159.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.159.0
└──▷ TRY IT
Configure enterprise credentials and parameters without manually editing config files.
copy
$ crewai enterprise configure
Adds enterprise configure CLI command for streamlined enterprise setup. Not checked yet Introduces partial flow resumability, enabling flows to resume from an intermediate state rather than restarting entirely. Not checked yet
2 details in Everything
CrewAI 0.157.0 adds a crewai config CLI command group, Okta device authorization, LangDB integration, and initial tracing capabilities.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.157.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.157.0
└──▷ TRY IT
Inspect or set CrewAI CLI configuration without manually editing config files.
Adds crewai config CLI command group for managing CLI configuration. Not checked yet Supports device authorization with Okta for authenticated flows. Not checked yet Adds LangDB integration for observability and query tracing. Not checked yet Introduces initial tracing capabilities for crew and flow execution. Not checked yet Enables persisting Flow state with BaseModel entries. Not checked yet
+1 more
show less
Adds default value support for crew.name. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
! Support for the deprecated User Memory system has been dropped; any setup relying on it will stop working after upgrading.
CrewAI 0.152.0 adds custom Flow names, a dedicated RAG RAG Retrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. module, and timezone-aware event timestamps.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.152.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.152.0
Supports custom flow names on the Flow class for clearer identification of flows. Not checked yet Refactors RAG components into a dedicated top-level module for cleaner imports and organization. Not checked yet Adds timezone support to event timestamps for accurate time-based event tracking. Not checked yet
3 details in Everything
CrewAI 0.150.0 adds ad-hoc tool calling tool calling A model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. , Mem0 v2 storage, SerperScrapeWebsiteTool, and Bedrock AgentCore browser/code interpreter toolkits.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.150.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.150.0
Adds support for ad-hoc tool calling via the internal LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. class, enabling on-demand tool invocation outside of standard crew/agent flows. Not checked yet Upgrades Mem0 Storage integration from v1.1 to v2. Not checked yet New SerperScrapeWebsiteTool extracts clean content from URLs using Serper. Not checked yet Integrates Bedrock AgentCore browser and code interpreter toolkits for use with Bedrock agents. Not checked yet Adds UserMemory deprecation notice to signal future removal. Not checked yet
5 details in Everything
CrewAI 0.148.0 introduces Agent evaluation functionality with thread-safe AgentEvaluator and neatlogs integration.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.148.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.148.0
Introduces Agent evaluation functionality with AgentEvaluator, supporting regression testing and experiment methods for both Agent and LiteAgent. Not checked yet Enables event emission during Agent evaluation for observability into evaluation runs. Not checked yet Adds crew context tracking for LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. guardrail events. Not checked yet Adds integration with neatlogs for structured agent log management. Not checked yet
4 details in Everything
CrewAI 0.141.0 adds crew context tracking for LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. guardrail events.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.141.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.141.0
CrewAI 0.140.0 adds LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. call tracking by task/agent, MemoryEvents monitoring, and a WorkOS CLI login command.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.140.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.140.0
Tracks LLM calls broken down by task and agent for observability into crew execution costs and patterns. Not checked yet Introduces MemoryEvents to monitor memory usage within a crew run. Not checked yet Adds console logging for memory system and LLM guardrail events. Not checked yet Improves data training support for models up to 7B parameters. Not checked yet Adds workos login command to the CLI for WorkOS-based authentication. Not checked yet
5 details in Everything
CrewAI 0.134.0 adds MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. multi-tool agent support, Tool-attribute initialization, and Oxylabs web scraping tools.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.134.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.134.0
Supports initializing a tool directly from defined Tool attributes, enabling programmatic tool construction. Not checked yet Adds an official way to use MCP Tools within a CrewBase class. Not checked yet Enhances MCP tools support to allow selecting multiple tools per agent inside CrewBase. Not checked yet Adds Oxylabs Web Scraping tools as a built-in integration. Not checked yet
4 details in Everything
CrewAI 0.130.0 adds LiteAgent with Guardrail integration, async tool execution, and multi-org CLI support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.130.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.130.0
Introduces LiteAgent with built-in Guardrail integration for lightweight, constrained agent workflows. Not checked yet Enables async tool execution for more efficient, non-blocking agent workflows. Not checked yet Adds support for multi-org actions in the CLI. Not checked yet Upgrades LiteLLM to support the latest OpenAI version. Not checked yet
4 details in Everything
CrewAI 0.126.0 adds Python 3.13 support, streamable-HTTP MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. transport, prompt/memory transparency, and tool-logging.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.126.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.126.0
Adds Python 3.13 support. Not checked yet Adds streamable-HTTP transport support in MCP integration. Not checked yet Enables tools to be loaded from an Agent repository via their own module. Not checked yet Persists available tools from a Tool repository across sessions. Not checked yet Logs tool usage when called by an LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. for observability. Not checked yet
+2 more
show less
Introduces transparency features for prompts and memory systems. Not checked yet Adds community analytics support. Not checked yet
7 details in Everything
CrewAI 0.121.0 adds markdown rendering for Tasks, reasoning for Agents, automatic date injection, and a HallucinationGuardrail.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.121.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.121.0
└──▷ USE IT
Enable markdown output for a task so results are returned as formatted markdown.
python
copy
from crewai import Task
task = Task(
description="Summarize the latest threat intelligence report.",
expected_output="A structured summary of key findings.",
markdown=True
)
Enable reasoning on an agent and inject today's date automatically for time-sensitive analysis workflows.
python
copy
from crewai import Agent
analyst = Agent(
role="Threat Analyst",
goal="Identify emerging threats from recent feeds.",
backstory="Expert in cyber threat intelligence.",
reasoning=True,
inject_date=True
)
Adds markdown attribute to the Task class for controlling markdown-formatted output. Not checked yet Adds reasoning attribute to the Agent class to enable or configure agent reasoning behavior. Not checked yet Adds inject_date flag to Agent for automatic date injection into agent context. Not checked yet Implements HallucinationGuardrail for detecting and guarding against hallucinated outputs. Not checked yet
4 details in Everything
CrewAI 0.120.0 adds agent-from-repository loading, empty Task context, and direct knowledge initialization.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.120.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.120.0
Supports loading an Agent directly from a repository. Not checked yet Enables setting an empty context for a Task. Not checked yet Introduces direct initialization of knowledge, bypassing knowledge_sources. Not checked yet
3 details in Everything
CrewAI 0.119.0 adds parent flow identification for Crew and LiteAgent and knowledge retrieval prompt rewriting in Agent.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.119.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.119.0
Enables parent flow identification for Crew and LiteAgent, improving traceability in nested flow architectures. Not checked yet Introduces knowledge retrieval prompt rewriting in Agent for improved tracking and debugging of RAG RAG Retrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. -based workflows. Not checked yet
2 details in Everything
CrewAI 0.118.0 adds no-code Guardrail creation and renames TaskGuardrail to LLMGuardrail.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.118.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.118.0
Adds support for no-code Guardrail creation to simplify AI behavior controls without writing custom guardrail logic. Not checked yet
1 detail in Everything
└──▷ BREAKING ON UPGRADE
! TaskGuardrail is renamed to LLMGuardrail ; any code importing or referencing TaskGuardrail will break on upgrade.
CrewAI 0.117.0 adds result_as_answer decorator support, GPT-4.1/Gemini-2.x models, and a HuggingFace CLI provider.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.117.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.117.0
└──▷ USE IT
Return a tool's output directly as the agent's final answer, bypassing further LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. reasoning — useful for deterministic lookup tools where you trust the result completely.
python
copy
from crewai.tools import tool
@tool(result_as_answer=True)
def lookup_cve(cve_id: str) -> str:
"""Fetch CVE details from internal database."""
return fetch_cve_record(cve_id)
Adds result_as_answer parameter to the @tool decorator, allowing a tool's output to be used directly as the agent's final answer. Not checked yet Supports new language models: GPT-4.1, Gemini-2.0, and Gemini-2.5 Pro. Not checked yet Adds HuggingFace as a provider option in the CrewAI CLI. Not checked yet Enhances knowledge management capabilities. Not checked yet
4 details in Everything
CrewAI 0.114.0 lets agents run standalone, adds custom LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. support, external memory, and Opik observability.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.114.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.114.0
└──▷ USE IT
Run a single agent as a standalone unit — useful in Flows or one-off tasks without assembling a full Crew.
python
copy
from crewai import Agent
researcher = Agent(
role="Research Analyst",
goal="Find the latest CVEs for Apache HTTP Server",
backstory="You are an expert in vulnerability research.",
llm="gpt-4o"
)
result = researcher.kickoff()
print(result)
Enables agents as atomic, standalone units — call Agent(...) .kickoff() without a full Crew. Not checked yet Supports custom LLM implementations for bringing your own model client. Not checked yet Integrates External Memory for persistent agent knowledge across runs. Not checked yet Adds Opik observability integration for tracing and monitoring agent workflows. Not checked yet Adds wildcard support to emit() for flexible event broadcasting. Not checked yet
+3 more
show less
Introduces secure fingerprints for agents and crews to uniquely identify and track them. Not checked yet Adds multimodal multimodal Able to work with more than one type of input or output, such as text, images and audio. A multimodal model can combine information across those formats. agent validation to enforce correct configuration of multimodal agents. Not checked yet Enhanced YAML extraction for more robust crew and agent definition parsing. Not checked yet
8 details in Everything
CrewAI 0.108.0 adds agent fingerprints, richer event listener visualization, and improved LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. streaming event handling.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.108.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.108.0
Adds fingerprints to agents/tasks for unique identity tracking across runs. Not checked yet Enhances LLM streaming response handling with an improved event system for real-time observability. Not checked yet Enriches the event listener with rich visualization and improved logging output. Not checked yet Includes model_name in relevant model representations for better introspection. Not checked yet
4 details in Everything
CrewAI 0.105.0 adds Flow state export, an event emitter for LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. observability, multi-router support, and Python 3.10 compatibility.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.105.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.105.0
Adds Flow state export and improved state utilities for inspecting and persisting Flow execution state. Not checked yet Introduces an event emitter for observability, enabling tracking of LLM calls and agent events. Not checked yet Supports multiple router calls within a single Flow, enabling more complex routing logic. Not checked yet Adds support for Python 3.10. Not checked yet Adds ChatOllama integration via langchain_ollama for local LLM usage. Not checked yet
+3 more
show less
Adds context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. size support for the o3-mini model. Not checked yet Enhances agent knowledge setup with an optional crew-level embedder configuration. Not checked yet Adds QdrantVectorSearchTool guide and event listener usage documentation. Not checked yet
8 details in Everything
CrewAI 0.102.0 adds QdrantVectorSearchTool, JSON logging, multi-tab Excel knowledge, and custom embedder support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.102.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.102.0
└──▷ USE IT
Search a Qdrant vector collection from within a CrewAI agent tool to ground responses in your own embeddings embeddings Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. .
python
copy
from crewai_tools import QdrantVectorSearchTool
tool = QdrantVectorSearchTool(
collection_name="security-advisories",
url="http://localhost:6333",
api_key="<your-qdrant-api-key>"
)
agent = Agent(role="Threat Analyst", tools=[tool], ...)
Adds QdrantVectorSearchTool for vector similarity search against Qdrant collections. Not checked yet Supports JSON format for logging output, improving observability pipeline integration. Not checked yet Enables multi-tab Excel file processing in excel_knowledge_source.py for richer knowledge ingestion. Not checked yet Adds a reset_memories function to the Crew class for programmatic memory management. Not checked yet Supports custom embedder configuration for enhanced embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. setup in knowledge sources. Not checked yet
+1 more
show less
Integrates MLflow tracing support for agent execution observability. Not checked yet
6 details in Everything
CrewAI 0.100.0 adds Amazon SageMaker as a supported LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.100.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.100.0
Supports Amazon SageMaker as an LLM provider for running agents against hosted SageMaker endpoints. Not checked yet
1 detail in Everything
CrewAI 0.98.0 adds Conversation Crew, flow state persistence with @persist, and SambaNova/NVIDIA NIM/VoyageAI integrations.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.98.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.98.0
└──▷ USE IT
Persist flow state across runs so a long-running security workflow can resume where it left off after interruption.
python
copy
from crewai.flow.persistence import persist, FlowPersistence
class MySecurityFlow(Flow):
@persist
def analyze_targets(self):
# state is automatically saved after this method completes
...
New Conversation Crew v1 enables interactive, dialogue-driven crew execution. Not checked yet Adds unique IDs to flow states for tracking and referencing individual flow runs. Not checked yet New @persist decorator with FlowPersistence interface enables durable flow state persistence across runs. Not checked yet Adds SambaNova as a new LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. provider integration. Not checked yet Adds NVIDIA NIM as a new provider via the CrewAI CLI. Not checked yet
+1 more
show less
Introduces VoyageAI as a new embedding embedding Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. /model integration. Not checked yet
6 details in Everything
CrewAI 0.95.0 adds multimodal multimodal Able to work with more than one type of input or output, such as text, images and audio. A multimodal model can combine information across those formats. agents, programmatic guardrails, multi-round HITL HITL Human in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. , Gemini 2.0, Langfuse, Portkey, Docling, and Weaviate support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.95.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.95.0
Adds multimodal abilities to Crew, enabling agents to process image and other non-text inputs. Not checked yet Introduces programmatic guardrails for enforcing constraints on agent outputs at runtime. Not checked yet Supports multiple rounds of Human-in-the-Loop (HITL) interaction within a single crew run. Not checked yet Adds Gemini 2.0 model support. Not checked yet Delivers CrewAI Flows improvements for more capable workflow orchestration. Not checked yet
+6 more
show less
Adds workflow permissions to control agent and task access within a crew. Not checked yet Supports Langfuse observability via LiteLLM integration. Not checked yet Adds Portkey integration for LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. gateway and observability. Not checked yet Introduces interpolate_only method on prompt/template handling for targeted variable substitution. Not checked yet Adds Docling support for document ingestion and parsing. Not checked yet Adds Weaviate support as a vector store vector store A database built to store embeddings and find the nearest ones to a query quickly. It is the retrieval half of most RAG systems. for agent knowledge. Not checked yet
11 details in Everything
CrewAI 0.86.0 adds multi-round Human-in-the-Loop follow-up and expanded knowledge tooling
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.86.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.86.0
Supports multiple rounds of Human-in-the-Loop (HITL HITL Human in the loop, involving a person in an otherwise automated workflow. The person can review, approve, correct or supply information before work continues. ) follow-up, enabling iterative human feedback within a crew's execution flow. Not checked yet Adds Nvidia NIM as a supported provider for custom LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. configuration. Not checked yet Introduces a knowledge demo and improved knowledge documentation to help practitioners integrate knowledge sources into crews. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! All references to Pipeline and PipelineRouter have been removed; any working setup that uses these constructs will break on upgrade.
CrewAI 0.85.0 adds agent-level knowledge, removes LangChain dependency, and improves typed task outputs.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.85.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.85.0
Adds knowledge support at the individual agent level, enabling per-agent knowledge bases alongside crew-level knowledge. Not checked yet Removes LangChain as a dependency, reducing the library's footprint and eliminating LangChain version conflicts. Not checked yet Improves typed task outputs for stronger, more predictable structured results from tasks. Not checked yet Adds Tool Repository authentication via crewai login, enabling access to hosted tools. Not checked yet
4 details in Everything
└──▷ BREAKING ON UPGRADE
! LangChain has been removed as a dependency; any code or configuration that imports or relies on LangChain internals through CrewAI will break on upgrade.
CrewAI v0.83.0 adds crew lifecycle callbacks, agent knowledge pre-seeding, and Mem0 memory/preference retrieval.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.83.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.83.0
└──▷ USE IT
Run setup or teardown logic around a crew execution using the new lifecycle hooks.
python
copy
@crew
def my_crew(self) -> Crew:
return Crew(
agents=self.agents,
tasks=self.tasks,
before_kickoff=self.before_kickoff,
after_kickoff=self.after_kickoff,
)
Adds before_kickoff and after_kickoff crew callbacks for hooking into crew lifecycle events. Not checked yet Supports pre-seeding agents with Knowledge so agents start with domain context before execution begins. Not checked yet Adds Mem0 integration for retrieving user preferences and memories during agent runs. Not checked yet
3 details in Everything
CrewAI 0.79.0 adds flow inputs, IBM Watson memory integration, and broader log storage data type support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.79.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.79.0
└──▷ USE IT
Build a custom tool using the now top-level BaseTool import rather than reaching into submodules.
python
copy
from crewai import BaseTool
class MyScanner(BaseTool):
name: str = "Port Scanner"
description: str = "Scans open ports on a given host."
def _run(self, host: str) -> str:
# tool logic here
return f"Scanning {host}"
Adds inputs parameter support to flows, enabling dynamic data to be passed into flow execution. Not checked yet Enhances log storage to support a wider range of data types beyond strings. Not checked yet Moves BaseTool to the main package and centralizes tool description generation, simplifying custom tool authoring. Not checked yet Raises an explicit error when an LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. returns no response, making silent failures visible. Not checked yet
4 details in Everything
CrewAI 0.76.9 adds crewai flow add-crew and updates the flow plot command to crewai flow plot.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.76.9 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.76.9
└──▷ TRY IT
Add a new crew to an existing flow without manually editing boilerplate — useful when extending a flow with a parallel or sequential crew.
copy
$ crewai flow add-crew
Visualize your flow's structure to review crew connections and routing logic before deploying.
New crewai flow add-crew CLI command lets you scaffold and add additional crews to an existing flow. Not checked yet Flow visualization command updated to crewai flow plot. Not checked yet Forwards install command options to uv sync for more flexible dependency management. Not checked yet Python 3.10 support added via tomli dependency. Not checked yet
4 details in Everything
└──▷ BREAKING ON UPGRADE
! The flow plot command is renamed from its previous form to crewai flow plot; any scripts or docs referencing the old command will break.
CrewAI 0.76.0 adds unsafe code execution support with Docker install and runtime checks.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.76.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.76.0
CrewAI 0.74.0 adds model selection and API key submission via CLI, a new Memory Base, and migrates tooling to UV.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.74.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.74.0
Migrates the project toolchain from Poetry to UV, with CLI adapted for UV workflows. Not checked yet New CLI capability to select models and submit API keys directly from the command line. Not checked yet Introduces a new Memory Base foundation for agent memory management. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! Project toolchain has migrated from Poetry to UV; existing Poetry-based setups will require migration to continue working.
CrewAI 0.70.1 adds Flows with a visual debugger, plus new CLI commands for scaffolding flows and tools.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.70.1 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.70.1
└──▷ TRY IT
Scaffold a new Flow project to start building multi-step agentic pipelines.
copy
$ crewai create flow
Scaffold a new custom tool ready for implementation and publishing.
copy
$ crewai tool create <tool>
New Flow feature enables structured, multi-step agentic pipelines. Not checked yet Flow visualizer lets practitioners inspect and debug Flow execution graphs. Not checked yet New crewai create flow command scaffolds a Flow project from the CLI. Not checked yet New crewai tool create <tool> command scaffolds a custom tool from the CLI. Not checked yet Adds Git validations when publishing tools to enforce clean repository state. Not checked yet
5 details in Everything
CrewAI 0.64.0 introduces an initial Tools API and raises the default max iterations to 20.
└──▷ GET THIS VERSION
copy
$ git clone --branch 0.64.0 https://github.com/crewAIInc/crewAI.git
$ git checkout 0.64.0
Adds an initial Tools API for programmatic tool management. Not checked yet Increases default max iterations from its previous limit to 20, enabling more complex agent reasoning loops out of the box. Not checked yet
2 details in Everything
CrewAI v0.63.0 adds a unified LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. class, custom memory interfaces, and switches the default model to GPT-4o-mini.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.63.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.63.0
└──▷ USE IT
Instantiate an agent with the new LLM class to explicitly target a specific model via LiteLLM, decoupling model selection from environment defaults.
python
copy
from crewai import Agent, LLM
llm = LLM(model="gpt-4o")
agent = Agent(
role="Security Analyst",
goal="Identify vulnerabilities in the provided code.",
backstory="Expert in application security.",
llm=llm
)
New LLM class provides a unified interface for interacting with language models via LiteLLM. Not checked yet Supports custom memory interfaces, enabling practitioners to plug in their own memory backends. Not checked yet Changes the default model to GPT-4o-mini. Not checked yet
3 details in Everything
└──▷ BREAKING ON UPGRADE
! The default model is now GPT-4o-mini; crews that relied on the previous default model will use GPT-4o-mini after upgrading unless an explicit model is set.
CrewAI v0.60.0 drops LangChain, rebuilds the executor, and adds o1-model support with new agent controls.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.60.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.60.0
└──▷ USE IT
Use CrewAI with an o1-series model that doesn't support system prompts or stop words.
python
copy
from crewai import Agent
agent = Agent(
role="Analyst",
goal="Analyze the dataset",
backstory="You are a data expert.",
llm="o1-preview",
use_system_prompt=False,
use_stop_words=False
)
Cap API usage and prevent runaway agent loops by setting rate limits and a max iteration count.
python
copy
from crewai import Agent
agent = Agent(
role="Researcher",
goal="Find key insights",
backstory="You are a senior researcher.",
max_rpm=10,
max_iter=5
)
Removes LangChain dependency with a fully rebuilt internal executor for improved reliability. Not checked yet New use_system_prompt flag on Agent lets you disable system prompts for models that don't support them. Not checked yet New use_stop_words flag on Agent allows disabling stop words to support o1-series models. Not checked yet Adds configurable max requests per minute at the crew/agent level. Not checked yet Adds a configurable maximum number of iterations before an agent is forced to produce a final answer. Not checked yet
+3 more
show less
New token calculation flow for accurate usage tracking across agent runs. Not checked yet New logging of crew and agent execution for improved observability. Not checked yet sliding_context_window is renamed to respect_context_window and is now enabled by default. Not checked yet
8 details in Everything
└──▷ BREAKING ON UPGRADE
! The sliding_context_window setting is renamed to respect_context_window; any existing config or code referencing sliding_context_window will break.! Delegation is now disabled by default; crews that relied on agents delegating tasks without explicit configuration will stop delegating on upgrade.
CrewAI v0.55.2 adds auto-complete, enriched TaskOutput fields, new install/deploy CLIs, and Pipeline cleanup.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.55.2 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.55.2
└──▷ TRY IT
Set up a new CrewAI project and its dependencies in one step.
Deploy a CrewAI project to a hosted environment after building it locally.
Inspect the name and expected output of a completed task result for downstream processing or logging.
python
copy
result = crew.kickoff()
for task_output in result.tasks_output:
print(task_output.name, task_output.expected_output)
Adds auto-complete support for CrewAI inputs. Not checked yet Adds name and expected_output fields to TaskOutput for richer task result inspection. Not checked yet New crewai install CLI command for streamlined project setup. Not checked yet New crewai deploy CLI command to deploy CrewAI projects. Not checked yet Cleans up and stabilizes the Pipeline feature. Not checked yet
5 details in Everything
CrewAI v0.51.0 adds crew testing/evaluation, pipelines, four new tools (Vision, DALL-E, MySQL, NL2SQL), and a sliding context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. .
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.51.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.51.0
└──▷ USE IT
Assign a cheaper or faster LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. specifically to the planning step so your expensive model is reserved for agent execution.
python
copy
from crewai import Crew, Agent, Task, Process
from langchain_openai import ChatOpenAI
planner_llm = ChatOpenAI(model="gpt-4o-mini")
crew = Crew(
agents=[...],
tasks=[...],
process=Process.sequential,
planning=True,
planning_llm=planner_llm
)
crew.kickoff()
Give agents the ability to generate images on demand during a workflow using the new DALL-E Tool.
python
copy
from crewai_tools import DallETool
from crewai import Agent
image_agent = Agent(
role="Image Creator",
goal="Generate visuals from descriptions",
backstory="You create images for marketing campaigns.",
tools=[DallETool()]
)
Enable natural-language database querying so agents can answer data questions without writing SQL manually.
python
copy
from crewai_tools import NL2SQLTool
from crewai import Agent
db_agent = Agent(
role="Data Analyst",
goal="Answer business questions from the database",
backstory="You query databases using plain English.",
tools=[NL2SQLTool(db_uri="mysql+pymysql://user:pass@host/dbname")]
)
Adds crew testing and evaluation framework for assessing crew performance. Not checked yet Introduces Pipeline structure for composing multi-crew workflows. Not checked yet Adds sliding context window to manage long-running agent context automatically. Not checked yet Supports setting a dedicated LLM for the planning step, separate from agent LLMs LLMs A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. . Not checked yet New crew run CLI command for executing CrewAI projects. Not checked yet
+8 more
show less
Allows all agent/task attributes to be defined in YAML project configuration. Not checked yet New Vision Tool for enabling agents to process and reason about images. Not checked yet New DALL-E Tool for generating images from within agent workflows. Not checked yet New MySQL Tool for querying MySQL databases from agents. Not checked yet New NL2SQL Tool for translating natural language queries into SQL. Not checked yet File-saving now serializes dict outputs as JSON when writing to disk. Not checked yet Enables verbose settings for tool outputs to aid in debugging and observability. Not checked yet Adds new GitHub project templates for bootstrapping CrewAI projects. Not checked yet
13 details in Everything
CrewAI v0.41.0 adds crew planning, task replay, memory reset, LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. retry, and type-safe outputs.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.41.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.41.0
└──▷ USE IT
Enable pre-execution planning so your crew deliberates a strategy before assigning tasks — useful for complex, multi-step investigations.
python
copy
from crewai import Crew
crew = Crew(
agents=[...],
tasks=[...],
planning=True
)
result = crew.kickoff()
Replay from a specific failed task after a partial run, without restarting the entire crew from scratch.
copy
$ crewai replay <task_id>
Clear stale crew memory before re-running to ensure the crew starts fresh without prior context contaminating results.
copy
$ crewai reset-memory
Adds planning=True to Crew instances so crews reason through a plan before executing tasks. Not checked yet Introduces a CLI replay feature to list tasks from the last run and re-execute from a specific task. Not checked yet Enables resetting crew memory before a run via a new reset-memory capability. Not checked yet Adds LLM call retry support so a failed LLM call no longer halts crew execution. Not checked yet All crews and tasks now return typed CrewOutput and TaskOutput objects instead of raw strings. Not checked yet
+3 more
show less
Adds ability to customize the output converter on agents/tasks. Not checked yet Enhances tools with type hinting and new attributes. Not checked yet Adds MultiON Tool integration. Not checked yet
8 details in Everything
└──▷ BREAKING ON UPGRADE
! All crews and tasks now return TaskOutput and CrewOutput objects instead of plain strings — code that treats crew/task return values as strings will break.
CrewAI v0.36.0 adds AgentOps native support, Firecrawl Tools, and tool-result-as-agent-result capability
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.36.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.36.0
Adds native AgentOps integration for agent observability and monitoring. Not checked yet Adds Firecrawl Tools for web scraping and crawling within CrewAI agents. Not checked yet Adds the ability to return a tool's result directly as an agent result. Not checked yet Improves the coding Interpreter tool. Not checked yet Adds the ability to create a custom converter class. Not checked yet
5 details in Everything
CrewAI v0.35.0 adds code execution for agents, third-party agent integration, and a new crewai train CLI command.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.35.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.35.0
└──▷ TRY IT
Train your crew for a fixed number of iterations before a production run to improve output consistency.
New crewai train -n <X> CLI command lets you train a crew for a specified number of iterations before execution to produce more consistent results. Not checked yet Agents can now execute code directly as part of a crew workflow. Not checked yet Supports integrating third-party agents — including LlamaIndex, LangChain, and Autogen agents — as first-class crew members. Not checked yet
3 details in Everything
CrewAI v0.32.0 adds async and per-item kickoff methods, LlamaIndex hub support, and usage metrics on crew output.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.32.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.32.0
└──▷ USE IT
Run the same crew independently for each item in a list — useful for processing multiple targets in parallel without managing async yourself.
python
copy
results = my_crew.kickoff_for_each(inputs=[{"target": "host1"}, {"target": "host2"}, {"target": "host3"}])
Kick off a long-running crew without blocking — lets you launch a crew and continue other work while it runs.
python
copy
import asyncio
async def main():
result = await my_crew.kickoff_async(inputs={"target": "host1"})
print(result)
asyncio.run(main())
Inspect token consumption after a crew run to track LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. costs per execution.
python
copy
result = my_crew.kickoff(inputs={"target": "host1"})
print(result.usage_metrics)
Adds kickoff_for_each, kickoff_async, and kickoff_for_each_async methods to the crew kickoff API for parallel and per-item execution control. Not checked yet Adds usage_metrics field to full crew output, exposing token and resource consumption data. Not checked yet Adds support for all LlamaIndex hub integrations as crew tools. Not checked yet Adds support for multiple crews in the new YAML format. Not checked yet Changes the default LLM model to gpt-4o. Not checked yet
+1 more
show less
Adds timestamps to log output. Not checked yet
6 details in Everything
└──▷ BREAKING ON UPGRADE
! The default model is now gpt-4o; crews that relied on the previous default model will switch automatically on upgrade.
CrewAI v0.30.4 adds manager agent override, prompt/response templates for OSS OSS Open-source software: code whose licence permits anyone to inspect, modify, and redistribute it. Cyber tools build on it to reuse auditable libraries without licensing fees. models, and Browserbase and Exa Search tools.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.30.4 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.30.4
Adds ability to designate a specific agent as crew manager instead of having the crew auto-generate one. Not checked yet Adds system, prompt, and response templates so practitioners can tune LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. interaction for open-source and smaller models. Not checked yet Adds initial support for bringing your own prompts to override built-in crew prompts. Not checked yet Adds two new built-in tools: Browserbase and Exa Search. Not checked yet Improves JSON and Pydantic output handling for better compatibility with smaller models. Not checked yet
+2 more
show less
Improves tool name recognition for better compatibility with smaller models. Not checked yet Adds ability to automatically create a directory when saving output as a file. Not checked yet
7 details in Everything
└──▷ BREAKING ON UPGRADE
! Dependencies have been updated — verify your tool integrations after upgrading.
CrewAI v0.27.0 adds shared crew memory, native human input, universal RAG RAG Retrieval-augmented generation: looking up relevant documents and adding them to the prompt before the model answers. It grounds responses in private or current data without retraining. tool support, and custom cache control.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.27.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.27.0
└──▷ USE IT
Enable shared crew memory so agents retain context across tasks, improving consistency in multi-step workflows.
python
copy
crew = Crew(
agents=[researcher, writer],
tasks=[research_task, write_task],
memory=True
)
Use a custom cache function on a tool to control exactly when results are cached, e.g. skipping cache for volatile data.
python
copy
from crewai_tools import tool
@tool
def my_tool(query: str) -> str:
...
my_tool.cache_function = lambda args, result: 'volatile' not in args['query']
Adds memory=True parameter to crew configuration to enable shared crew memory, improving outcome reliability (disabled by default). Not checked yet Adds cache_function attribute to tools for custom caching logic per tool invocation. Not checked yet Adds native human input support, allowing agents to pause and request input from a human during execution. Not checked yet Extends RAG tools support to any embedding model embedding model Lists of numbers a model produces to represent the meaning of text, images or code, so similar items sit close together. They power semantic search and RAG. and provider, no longer limited to OpenAI. Not checked yet Adds cross-agent delegation, enabling smoother cooperation and task handoff between agents. Not checked yet
5 details in Everything
CrewAI v0.22.0 adds a crewai create CLI command and dictionary-based agent/task definitions.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.22.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.22.0
└──▷ TRY IT
Bootstrap a new CrewAI project scaffold from the command line.
Define an agent using a dictionary instead of a class instance, useful for dynamic or config-driven crew setups.
python
copy
from crewai import Agent, Task, Crew
agent = Agent(**{
"role": "Researcher",
"goal": "Find the latest AI news",
"backstory": "You are an expert at finding information."
})
task = Task(**{
"description": "Search for the top 5 AI news stories today",
"agent": agent
})
Adds crewai create CLI command for bootstrapping new CrewAI projects from the command line. Not checked yet Enables agents and tasks to be defined using Python dictionaries as an alternative to class-based definitions. Not checked yet Adds clearer agent logging output to improve observability during crew execution. Not checked yet
3 details in Everything
CrewAI v0.19.0 adds execution metrics, input passing at kickoff, and function-calling function-calling A model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. fallback for agents and crews.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.19.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.19.0
└──▷ USE IT
After a crew run, inspect tool usage metrics to audit how many tools were called and where errors occurred.
python
copy
result = crew.kickoff(inputs={'domain': 'example.com'})
print(crew.usage_metrics)
Fall back to function-calling mode on an agent when the default tool-use strategy is unreliable with your LLM.
python
copy
agent = Agent(
role='Analyst',
llm=my_llm,
function_calling_llm=my_function_llm,
tools=[my_tool]
)
Adds function_calling_llm parameter on Agent or Crew to fall back to function calling when standard tool usage fails. Not checked yet Adds crew.usage_metrics to retrieve execution metrics after a kickoff call, exposing tool usage statistics. Not checked yet Adds inputs parameter to crew.kickoff(inputs={'key': 'value'}) so runtime values can be injected at execution time. Not checked yet Enhances delegation capabilities for agents within a crew. Not checked yet
4 details in Everything
CrewAI v0.16.0 adds inputs interpolation and telemetry for tools usage, errors, and token tracking.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.16.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.16.0
Adds initial support for inputs interpolation, enabling dynamic input substitution in crew workflows. Not checked yet Adds ability to track tools usage, tools errors, formatting errors, and token usage within crew runs. Not checked yet
2 details in Everything
└──▷ BREAKING ON UPGRADE
! The crewai_tools dependency has been removed; any setup relying on it will break on upgrade.
CrewAI v0.14.0rc0 adds crewai-tools integration, Pydantic/JSON task output formatting, and file output support.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.14.0rc0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.14.0rc0
Adds support for crewai-tools integration. Not checked yet Adds support for formatting task output as Pydantic objects or JSON. Not checked yet Adds support for saving task output to a file. Not checked yet Supports tools with no arguments. Not checked yet Revamps tools usage logic to properly use function calling function calling A model's ability to answer with a structured request to run a named function or API, which the application executes and returns. It is how agents take actions. . Not checked yet
+1 more
show less
Improves reliability for inter-agent delegation. Not checked yet
6 details in Everything
CrewAI v0.10.0 adds full task output capture, step callbacks, multi-argument tool support via JSON, and shared caching across agents.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.10.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.10.0
└──▷ USE IT
Stream every agent reasoning step in real time — useful for debugging long-running autonomous crews.
python
copy
def my_callback(step):
print(step)
agent = Agent(..., step_callback=my_callback)
crew = Crew(agents=[agent], ..., step_callback=my_callback)
Adds full_output return from crew kickoff, exposing all individual task outputs in a single result. Not checked yet Adds step_callback parameter for both Agents and Crews to receive all intermediate reasoning steps during execution. Not checked yet New tool usage internals now use JSON, unlocking support for tools that require multiple arguments. Not checked yet Rebuilt caching structure so multiple agents can share the same cache across a crew run. Not checked yet Adds opt-in sharing of complete crew run data with the crewAI team. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! Removes CrewAgentOutputParser — any code that imports or references it directly will break on upgrade.
CrewAI v0.5.0 adds task callbacks, hierarchical process support, task references, and parallel task execution.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.5.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.5.0
Adds task callbacks, enabling post-execution hooks on individual tasks. Not checked yet Adds support for a hierarchical process model for structuring agent workflows. Not checked yet Adds the ability to reference specific tasks from within another task. Not checked yet Adds parallel task execution, allowing multiple tasks to run concurrently. Not checked yet
4 details in Everything
CrewAI v0.1.32 adds per-agent iteration limits, RPM throttling, and initial i18n support
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.1.32 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.1.32
Adds ability to limit the maximum number of iterations for an agent, preventing runaway agent loops. Not checked yet Adds Request Per Minute (RPM) throttling configurable for both individual Agents and Crews. Not checked yet Adds initial internationalization (i18n) support with a Greek translation included. Not checked yet
3 details in Everything
CrewAI v0.1.14 adds tool caching, loop execution prevention, verbose logging levels, and pydantic v2 support.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.1.14 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.1.14
Adds verbose levels to the logger for finer-grained output control. Not checked yet Adds tool caching to avoid redundant tool calls during agent execution. Not checked yet Adds loop execution prevention to stop agents from cycling indefinitely. Not checked yet Expands delegation guidelines to give agents more precise rules for task hand-off. Not checked yet Updates support to pydantic v2. Not checked yet
5 details in Everything
└──▷ BREAKING ON UPGRADE
! Upgrades to pydantic v2, which may break existing model definitions that rely on pydantic v1 behaviour.
CrewAI v0.1.1 adds verbose mode for inspecting task execution in real time.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.1.1 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.1.1
Adds Crew verbose mode to inspect tasks as they are being executed. Not checked yet
1 detail in Everything
CrewAI v0.1.0 debuts a Python framework for orchestrating collaborative, role-based autonomous AI agents.
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.1.0 https://github.com/crewAIInc/crewAI.git
$ git checkout v0.1.0
└──▷ USE IT
Stand up a minimal crew where two specialized agents tackle a research-then-write task sequentially.
python
copy
from crewai import Agent, Task, Crew, Process
researcher = Agent(
role='Researcher',
goal='Find key facts about quantum computing',
tools=[search_tool]
)
writer = Agent(
role='Writer',
goal='Summarize research into a short briefing',
tools=[]
)
research_task = Task(description='Research recent quantum computing breakthroughs', tools=[search_tool])
write_task = Task(description='Write a 200-word executive summary', tools=[])
crew = Crew(
agents=[researcher, writer],
tasks=[research_task, write_task],
process=Process.sequential
)
result = crew.kickoff()
print(result)
Introduces the Agent class for defining autonomous agents with specific roles, goals, and assigned tools. Not checked yet Introduces the Task class for dynamically creating and assigning tasks with per-task tool specifications. Not checked yet Introduces the Crew class for grouping agents and coordinating collaborative workflows. Not checked yet Introduces the Process class supporting sequential task execution for organized, predictable agent pipelines. Not checked yet Enables inter-agent delegation, allowing agents to autonomously redistribute subtasks among team members at runtime. Not checked yet
5 details in Everything