CrewAI

This release introduces significant new features for flow management, including dynamic model pulling and inline skill definitions. It also addresses several bug fixes related to caching, dependency vulnerabilities, and flow execution.

This release introduces new features related to flow agents and bedrock extras, alongside a bug fix rejecting self-listening flow methods.

This release introduces new features like inline skill definitions and flow protocol support, alongside a fix for documentation navigation issues.

This release introduces new features like Git initialization for generated projects and explicit CrewAI project definitions. Several bugs were fixed, including issues with deployment links, JSON template rendering, and SSRF vulnerabilities.

This release introduces new telemetry tracking, enforces explicit project definitions, and fixes several bugs related to JSON rendering and security vulnerabilities.

This release introduces significant features for conversational flows, including CLI support and declarative loading, alongside various bug fixes and internal refactoring for better flow state management.

This release introduces support for declarative refs across flows and crews and resolves an issue with JSON schema flow state kickoff inputs.

This release introduces support for conversational flows in the CLI TUI and includes fixes for symlink path traversal and declarative flow definition validation.

This release introduces significant improvements to declarative flow loading and CLI experience, alongside better UX for crew execution and typed output schemas for tools.

This release introduces validation for Flow CEL expressions upon definition loading and adds support for single agent actions in Flow definitions. Documentation updates include a new Datadog integration guide.

This release introduces an optional if expression for each.do steps and resolves issues related to JSON crew functionality.

This release introduces significant features centered around FlowDefinition, enabling script/code block actions, crew actions, and 'each' composite actions. It also implements DMN mode support and introduces JSON-first crew projects.

This release introduces significant extensibility with pluggable backends and a new chat API for conversational flows. It also includes numerous bug fixes, particularly around state restoration and Snowflake integration, alongside performance improvements for imports.

This release primarily focuses on a bug fix related to snapshot restoration, ensuring live snapshots are not replayed as resumes.

This release introduces enhanced prompt handling and state management capabilities, alongside fixes for runtime state isolation and telemetry setup.

This release introduces significant architectural updates by migrating the @listen/@router runtime and adding pluggable default backends for core components.

This release focuses on bug fixes, including resolving CVEs and correcting an exposure issue on AgentExecutor, alongside a minor refactoring to update how '@start' reads definitions.

This release introduces support for conversational flow traces, surfaces detailed LLM event information, and refactors the Flow DSL into focused decorator modules. It also includes documentation updates for new guides and deployment methods.

This release introduces support for crew trained agents and the Snowflake Cortex LLM provider, alongside several bug fixes, notably improving CLI stability and Snowflake Claude integration.

This release introduces enhancements to environment variable handling in StdioTransport and planning configuration, alongside several bug fixes related to structured output leaks and checkpointing reliability. The Skills Repository has also been moved to an experimental status.

This release introduces enhancements to StdioTransport and planning configurations, alongside several bug fixes related to structured output leaks and checkpointing reliability. Documentation has also been updated for Agent Control Plane and checkpointing.

This release introduces the Skills Repository feature and enhances enterprise release note generation. It also includes several bug fixes, notably addressing a security issue by bumping the idna dependency.

This release introduces improvements to sandbox tools and kickoff parameters, deprecates `CrewAgentExecutor` in favor of `AgentExecutor`, and fixes several bugs including a memory leak in `git.py`.

This release primarily focuses on documentation updates and introduces a breaking change by deprecating the "function_calling_llm" field.

This release primarily focuses on bug fixes, including resolving an issue with streamed tool calls and updating the langsmith dependency to address a security vulnerability. Documentation for TavilyGetResearch was also added.

This release introduces improvements to Daytona sandbox tools and deprecates CrewAgentExecutor in favor of AgentExecutor. Several security patches and documentation updates were also included.

This release primarily updates LLM listings and resolves a dependency conflict by adjusting package placements.

This release fixes an issue with the status endpoint path and updates the gitpython dependency for security. The CLI has been extracted into a new standalone package.

This release focuses on bug fixes, improving output handling in various scenarios, and ensuring correct token counting. Key fixes involve output restoration, async flushing, and preventing incorrect final answers.

This release introduces a new kickoff parameter 'restore_from_state_id' and updates the ExaSearchTool naming convention. Several minor bug fixes related to release flow and tracing were also implemented.

This release introduces several new features, including custom persistence key support and expanded search/research tools via Tavily and You.com. Numerous bug fixes address issues related to JSON handling, tool call preservation, and agent execution.

This release introduces support for custom persistence keys and enhances Azure OpenAI integration with Responses API support and credential forwarding. Several minor bug fixes address LLM handling and tool execution.

This release focuses on stability and security by fixing several bugs related to LLM interactions, executor state management, and dependency updates, including a bump to litellm for an SSTI fix. New documentation for various sandbox tools has also been added.

This release introduces significant new features, including checkpoint lifecycle events, e2b support, and Bedrock V4 integration. It also includes numerous bug fixes related to serialization, checkpointing, and dependency upgrades for security.

This release introduces support for e2b and implements a fallback to DefaultAzureCredential. It also includes a security fix by upgrading the lxml dependency.

This release introduces support for Bedrock V4 and Daytona sandbox tools, alongside new 'Build with AI' documentation features. Several bugs related to event propagation, serialization, and guardrail retries have also been fixed.

This release introduces checkpoint and fork support for standalone agents and includes several bug fixes related to Gemini streaming, checkpointing, and devtools.

This release introduces significant new features around checkpoint management, including resume, diff, and forking capabilities. It also addresses several bugs and patches security vulnerabilities across multiple dependencies.

This release primarily focuses on security fixes by updating dependencies python-multipart and pypdf, and resolves an issue with cyclic JSON schema handling in MCP tool resolution.

This release primarily updates the changelog and version number to v1.14.2a4.

This release introduces resume hints for devtools failure releases and addresses several bugs, including strict mode forwarding to Bedrock Converse API and a security vulnerability in pytest.

This release introduces a new deploy validation CLI and improves LLM initialization ergonomics, alongside several critical security and stability fixes.

This release introduces significant enhancements to checkpoint management, including a new TUI and forking capabilities, alongside improvements to LLM token tracking and tool security.

This release primarily focuses on bug fixes, including resolving an event emission issue after HITL resume and updating the cryptography dependency to patch a security vulnerability.

This release introduces an async checkpoint TUI browser and adds closing methods/context managers for streaming outputs. It also includes several bug fixes, notably resolving a CVE by bumping the transformers dependency.

This release introduces an async checkpoint TUI browser and improved streaming output management. It also includes several bug fixes, notably resolving a CVE by bumping the transformers library.

This release introduces significant features around checkpointing, including configuration and runtime state management, alongside important security fixes and dependency updates.

This release introduces significant features like runtime state checkpointing and a new SqliteProvider for storage. It also removes the CodeInterpreterTool and addresses a security vulnerability via a dependency bump.

This release primarily updates the changelog and version number to v1.14.0a2.

This release introduces a unified RuntimeState RootModel, enhances telemetry for events, and adds A2UI extension support. Several core components like Flow and LLM classes were refactored to use Pydantic BaseModels, and various bugs related to credentials, model handling, and RBAC were fixed.

This release introduces the A2UI extension with support for v0.8/v0.9 and fixes multimodal vision prefixes by including GPT-5 and o-series models.

This release focuses on documentation fixes and performance improvements by reducing framework overhead through lazy event bus implementation and skipping tracing when disabled.

This release primarily updates the changelog and version number to v1.13.0a4.

This release primarily updates the changelog and version information to v1.13.0a3.

This release introduces new features for emitting token usage data and publishing tool metadata. It also includes bug fixes for handling GPT-5.x models and significant internal refactoring converting core components to Pydantic BaseModels.

This release primarily updates the changelog and version number to v1.13.0a2.

This release focuses on internal improvements, including auto-updating the deployment test repository and enhancing enterprise release resilience and user experience.

This release focuses on maintenance, fixing issues in the documentation workflow and enterprise package publishing, and updating documentation with an RBAC matrix.

This release introduces the enterprise release phase for devtools and fixes a bug related to preserving method return values in human feedback flows.

This release introduces significant new features like agent skills, expanded OpenAI-compatible provider support, and Qdrant Edge storage. It also includes numerous bug fixes, particularly around type checking and memory handling.

This release introduces significant new features, including Qdrant Edge support, agent skills, and native support for numerous OpenAI-compatible providers. It also includes extensive bug fixes, particularly around mypy type checking and memory handling.

This release focuses on bug fixes, specifically addressing credential issues for trace batch pushes and resolving several bugs within the HITL flow system.

This release introduces the Qdrant Edge storage backend for the memory system and updates documentation.

This release introduces significant new features like native OpenAI-compatible provider support and agent skills, alongside extensive bug fixes, particularly resolving mypy errors across packages. The internal plumbing related to LiteLLM has also been decoupled.

This release introduces flow structure serialization and addresses several bugs, including security vulnerabilities and stability fixes for tool handling and configuration persistence.

This release primarily updates the changelog and version number to v1.11.0rc2.

This release focuses on security and stability by upgrading vulnerable dependencies and improving LLM response handling. Documentation for several tools and guides has also been updated.

This release introduces Plus API token authentication and implements the plan execute pattern, alongside fixing a critical code interpreter sandbox escape issue.

This release primarily focuses on fixing an issue by removing exclusive locks from read-only storage operations.

This release introduces a new release command for PyPI publishing and includes several bug fixes related to thread safety and ContextVar propagation.

This release introduces token-saving tool search capabilities and adds support for more Brave Search tools. It also includes several bug fixes addressing concurrency issues, tool result grouping, and dependency vulnerabilities.

This release upgrades the Gemini GenAI integration and addresses several bugs, including resolving critical security alerts by upgrading the pypdf dependency.

This release introduces asynchronous invocation support for step callback methods and implements lazy loading for heavy dependencies within the Memory module.

This release introduces significant feature enhancements, including improved tool resolution, JSON argument parsing, and user input handling in Flows. Numerous bug fixes address issues related to tool validation, schema handling, and flow persistence.

This release introduces significant architectural refactoring, including a New Unified Memory System and provider pattern extraction, alongside numerous bug fixes for flow tracking, event handling, and tool usage. It also adds features like version checking and improved HITL support.

This release introduces several new features including a2a liteagent, enhanced authentication, transport negotiation, and file support, alongside bug fixes for agent output handling and response model integration.

This release focuses on minor fixes, including ensuring the verbosity flag is applied and correcting issues related to response handling.

This release introduces before and after tool call hooks in CrewAgentExecutor and adds structured outputs and response_format support across providers, alongside a fix for tool-calling content handling.

This release introduces significant new features including structured outputs, response_format support, native multimodal file handling, and Keycloak SSO. Several bug fixes address issues related to Bedrock, Google Vertex API, and flow error management.

This release introduces significant features related to A2A (Agent-to-Agent) communication, including task execution utilities and server configuration. Several bug fixes address issues in flow execution and error handling.

This release introduces production-ready Flows and Crews architecture, featuring native async agent-to-agent communication and human-in-the-loop feedback mechanisms.

This release focuses on resolving connection issues and updating the API reference documentation.

This release introduces a --no-commit flag for the bump command, implements JSON schema for tool serialization, and provides several bug fixes related to async task execution and platform compatibility.

This release introduces comprehensive native async support across the entire framework, including flows, agents, tools, and memory. It also includes critical bug fixes for OpenTelemetry integration, token store deadlocks, and Anthropic tool support.

This release focuses on stability improvements across the LLM and RagTool modules, including fixes for async execution and configuration resetting.

This release introduces streaming support for flows and crews, adds Gemini 3 Pro Preview support, and includes several bug fixes for RAG tools and OpenAI parameter handling.

This is a hotfix release updating the crewAI version from 0.203.1 to 0.203.2.

This release introduces LLM call hooks for CrewAgentExecutor, expands data exposure in Task and Agent outputs, and improves tracing instrumentation and documentation.

This release focuses on bug fixes for agent iteration handling and LLM provider routing logic.

This release introduces first-class MCP support and LLM message interceptor hooks, while refactoring embedding configurations and fixing critical flow state and caching issues.

This release focuses on refactoring flow handling and tool documentation, specifically renaming embedding parameters and enhancing the Qdrant and Firecrawl tools.

This release introduces Datadog integration support and enables apps and MCPs within liteagent, alongside updated documentation for platform tool environment variables.

This release focuses on maintenance by updating default LLM models, improving logging, and refining flow visualization, while also removing the unused aisuite dependency.

This release introduces support for multiple LLM providers in InternalInstructor, adds a mypy plugin base, and improves typing for CrewBase alongside several bug fixes.

CrewAI reaches version 1.0.0, introducing enhanced guardrail handling for Agents and credential injection for tools, alongside several stability fixes for Flows and Docker environments.

This release focuses on enhancing LLM completion classes (Bedrock, Gemini, Anthropic) with better parameter support and refactoring the project module to a metaclass for improved typing.

This release focuses on enhancing the OpenAICompletion class, improving event bus thread safety, and fixing critical bugs related to input availability and JWT decoding.

This release introduces Bedrock LLM integration and enhances OpenAICompletion parameters while improving event bus thread safety. It also includes several bug fixes related to JWT decoding, input availability, and task handling.

This release focuses on core improvements, including a fix for tool repository credential injection and enhanced JWT validation with a 10-second leeway.