Project directory

Orchestration

Tools for coordinating agents and multistep work.

Pydantic AI

Pydantic AI

Pydantic AI builds Python agents with typed tool inputs, dependencies and validated outputs.

  • Runtime. The package requires Python 3.11 or later. The slim install lets applications select provider and integration extras. Source ↗
  • Model access. Direct model calls use provider credentials. Pydantic AI Gateway is a separate route with its own key and spending controls. Source ↗

Publisher update ·

Pydantic AI 2.71.0 ↗

Scheduled · awaiting first completed check

Paperclip

Paperclip

Paperclip coordinates AI agents through an organization chart, task board and budget controls.

  • Self-hosting. The MIT-licensed server and web UI run locally with Node.js 24.11 or later and embedded PostgreSQL. A production deployment can use an external PostgreSQL server. Source ↗
  • Agent runtimes. Each agent can use its own model and harness. Adapters connect local tools such as Codex and Claude Code or external agents over HTTP. Source ↗

Publisher update ·

Paperclip v2026.1117.0 ↗

Scheduled · awaiting first completed check

CrewAI

CrewAI

CrewAI combines role-based teams of AI agents with event-driven workflows in Python.

  • Runtime. The MIT-licensed framework runs on Python 3.10 through 3.13. Source ↗
  • Crews. Agents have distinct roles and goals and can delegate tasks within a crew. Source ↗

Publisher update ·

CrewAI 1.15.42 ↗

Scheduled · awaiting first completed check

Mastra

Mastra

Mastra is a TypeScript framework for tool-using agents and explicit workflow graphs.

  • License. The core is Apache-2.0. Code in ee directories is source-available and requires an enterprise license for production use. Source ↗
  • Model access. The model router reads credentials for the selected provider or gateway. Model access is arranged separately from core framework licensing. Source ↗

Publisher update ·

Mastra core 1.84.0 ↗

Scheduled · awaiting first completed check

Multica

Multica

Multica assigns coding agents to issues and tracks their work in a shared workspace.

  • License. The source-available license permits internal use within one organization. Third-party hosting and commercial embedding require a commercial license, and enterprise directories have separate commercial terms. Source ↗
  • Self-hosting. The platform supports Docker Compose or Helm with PostgreSQL. A separate daemon runs agent CLIs on the execution machine. Source ↗

Publisher update ·

Multica 0.11.0 ↗

Scheduled · awaiting first completed check

OpenAI Agents SDK coordinates Python agent runs with tools, handoffs and guardrails.

  • Runtime. The MIT-licensed Python package requires Python 3.10 or later. Source ↗
  • Model access. The SDK supports OpenAI Responses and Chat Completions plus other model providers. Hosted inference requires credentials for the selected provider. Source ↗

Publisher update ·

OpenAI Agents SDK for Python 0.25.0 ↗

Scheduled · awaiting first completed check

Langflow

Langflow

Langflow provides a visual editor for composing, testing and serving AI workflows.

  • Local installs. The MIT-licensed OSS package runs on Python 3.10 through 3.14. Desktop builds bundle their dependencies for macOS and Windows. Source ↗
  • Authoring. Flows connect visual components with editable Python implementations, run in a playground and export as JSON. Source ↗

Publisher update ·

Langflow 1.13.3 ↗

Scheduled · awaiting first completed check

Agno

Agno

Agno is a Python framework and self-hostable runtime for serving agents, teams and workflows.

  • Access. Agno is distributed under Apache-2.0 and can run on infrastructure you operate. Source ↗
  • Model access. The OpenAI model integration requires the OpenAI Python package and an API key for inference. Source ↗

Publisher update ·

Agno 3.2.1 ↗

Scheduled · awaiting first completed check

Paradev (formerly Superset)

Paradev (formerly Superset)

Paradev, formerly Superset, is a desktop workspace manager for running coding agents in parallel and reviewing their work.

  • Workspace isolation. Local project workspaces use Git worktrees with separate branches and terminals. Cloud workspaces use separate repository clones. Source ↗
  • Platform. The desktop app supports Intel and Apple Silicon Macs. The Linux x64 AppImage is experimental, and Windows support is unavailable. Source ↗

Publisher update ·

Paradev Desktop 2.0.9 ↗

Scheduled · awaiting first completed check

AutoGPT

AutoGPT

AutoGPT Platform provides a visual builder and runtime for composing and running reusable agent workflows.

  • Visual builder. Workflows connect reusable blocks and branches, with run inspection and a marketplace for starting from existing workflows. Source ↗
  • Hosted access. The hosted Platform is a paid service with model access built in. Self-hosted installations use operator-provided model credentials or a local model. Source ↗

Publisher update ·

AutoGPT Platform beta 0.11.3 ↗

Scheduled · awaiting first completed check

LangGraph

LangGraph

LangGraph runs stateful agent workflows as graphs with checkpoints and resumable execution.

  • Python package. The MIT-licensed langgraph package requires Python 3.10 or later. Source ↗
  • State and memory. Checkpointers save state by thread. Stores hold data shared across threads, such as user preferences. Source ↗

Publisher update ·

LangGraph 1.2.22 ↗

Scheduled · awaiting first completed check

Google Agent Development Kit

Google Agent Development Kit

Google Agent Development Kit for Python is a code-first framework for building and deploying agents and workflows.

  • Package. The Apache 2.0 Python package is published as google-adk. The 2.14.1 package requires Python 3.11 or later. Source ↗
  • Orchestration. The graph runtime supports routing, parallel branches and joins, retries, nested workflows and human input. Source ↗

Publisher update ·

Google ADK for Python 2.14.1 ↗

Scheduled · awaiting first completed check

Semantic Kernel

Semantic Kernel

Semantic Kernel is a Microsoft SDK for connecting model services to application functions, prompts and agent workflows.

  • Project status. Microsoft identifies Agent Framework as the successor and provides migration guides for Semantic Kernel users. Source ↗
  • Model access. The SDK is MIT-licensed. Its OpenAI and Azure OpenAI examples require separate API credentials and model access. Source ↗

Publisher update ·

Semantic Kernel .NET 1.82.0 ↗

Scheduled · awaiting first completed check

Microsoft Agent Framework provides SDKs for building agents and coordinating them with explicit workflows.

  • Runtime. Python and .NET both support agents and workflows. The Go SDK is in public preview and requires Go 1.26 or later. Source ↗
  • Graph execution. The .NET workflow builder connects typed executors with conditional edges, supports fan-out and fan-in, and checkpoints at superstep boundaries. Source ↗

Scheduled · awaiting first completed check

Agency Swarm

Agency Swarm

Agency Swarm extends the OpenAI Agents SDK with directed communication and state management for teams of Python agents.

  • Runtime. The MIT-licensed framework requires Python 3.12 or later. Source ↗
  • Communication. An agency declares which agents may initiate messages to each other. Agent tools can validate arguments with Pydantic. Source ↗

Publisher update ·

Agency Swarm v1.11.0 ↗

Scheduled · awaiting first completed check

smolagents

smolagents

smolagents is a Python framework for building tool-using agents that can express actions as code or structured tool calls.

  • Agent loops. CodeAgent emits Python actions, while ToolCallingAgent uses structured tool calls. Source ↗
  • Model access. Adapters support local Transformers models and hosted providers through Hugging Face Inference Providers, LiteLLM or OpenAI-compatible endpoints. Hosted model calls use the selected service’s credentials and terms. Source ↗

Publisher update ·

smolagents 1.26.0 ↗

Scheduled · awaiting first completed check

AutoGen

AutoGen

Microsoft describes AutoGen, its framework for multi-agent applications, as archived and no longer maintained.

  • Maintenance. Microsoft's project notice says maintenance has ended, including new features, bug fixes, security fixes and support. Source ↗
  • Python runtime. AgentChat requires Python 3.10 or later. Optional extensions supply model clients, and AutoGen Studio is distributed as a separate graphical interface. Source ↗

Publisher update ·

AutoGen Python 0.7.5 ↗

Scheduled · awaiting first completed check

MetaGPT

MetaGPT

MetaGPT coordinates role-based language model agents to produce software specifications and project files from a requirement.

  • Role handoffs. Roles subscribe to upstream message types, so a tester can react when a coding role publishes its output. Source ↗
  • Runtime. The documented Python range for v0.8.2 is 3.9 through 3.11. Source ↗

Publisher update ·

MetaGPT 0.8.2 ↗

Scheduled · awaiting first completed check

Dify

LangGenius

Dify provides visual workflows, model integrations and retrieval pipelines for building and serving AI applications.

  • Deployment options. Dify Cloud is a managed service with its own plans and allowances, and Community Edition can run on operator-managed infrastructure. Source ↗
  • License scope. The repository uses a modified Apache 2.0 license with conditions on multi-tenant operation and removal of frontend branding. Source ↗

Monitored

n8n

n8n

n8n is a visual automation runtime that connects application integrations, code, API calls and AI steps.

  • Deployment options. Self-hosted installs without a license key run Community Edition. Business and Enterprise self-hosted features require license keys, and n8n Cloud is a separate managed service. Source ↗
  • License scope. The Sustainable Use License permits internal business, personal and noncommercial use and restricts offering n8n itself as a hosted product to customers. Source ↗

Scheduled · awaiting first completed check