{"ok":true,"source":"tensorfeed.ai","lastUpdated":"2026-10-19","count":17,"frameworks":[{"id":"langchain","name":"LangChain","vendor":"LangChain Inc.","languages":["python","typescript"],"version":"1.4.x","released":"2022-10-22","license":"MIT","githubStarsK":148.1,"weeklyInstallsK":37227,"category":"agent-orchestration","features":["chains","tools","memory","tracing (LangSmith)","document loaders","500+ integrations"],"url":"https://www.langchain.com","github":"https://github.com/langchain-ai/langchain","notes":"The most-installed Python agent framework, now on the 1.x line and billed by its maintainers as \"the agent engineering platform.\" langchain-core and per-provider packages ship separately to reduce churn. Pairs tightly with LangGraph for stateful multi-step workflows and LangSmith for tracing."},{"id":"langgraph","name":"LangGraph","vendor":"LangChain Inc.","languages":["python","typescript"],"version":"1.2.x","released":"2024-01-17","license":"MIT","githubStarsK":43.5,"weeklyInstallsK":12266,"category":"agent-orchestration","features":["stateful graphs","checkpointing","human-in-the-loop","streaming","time-travel"],"url":"https://docs.langchain.com/oss/python/langgraph/overview","github":"https://github.com/langchain-ai/langgraph","notes":"Stateful agent graphs from the LangChain team, now past 1.0. The framework many production agents moved to because of first-class checkpointing, human-in-the-loop interrupts, and time-travel debugging. Draws about 12 million weekly PyPI installs as of October 2026."},{"id":"llamaindex","name":"LlamaIndex","vendor":"LlamaIndex Inc.","languages":["python","typescript"],"version":"0.14.x","released":"2022-11-22","license":"MIT","githubStarsK":52.6,"weeklyInstallsK":648,"category":"rag","features":["document loaders","indexes","query engines","agents","workflows","observability"],"url":"https://www.llamaindex.ai","github":"https://github.com/run-llama/llama_index","notes":"RAG-first framework, strong on document ingestion, indexing strategies, and query rewriting. The company now describes LlamaIndex as a document processing platform for AI, with LlamaParse and hosted services alongside the open-source library. Workflows is the agent-orchestration layer."},{"id":"microsoft-agent-framework","name":"Microsoft Agent Framework","vendor":"Microsoft","languages":["python","multi"],"version":"1.22.x","released":"2025-10-01","license":"MIT","githubStarsK":14.2,"weeklyInstallsK":46,"category":"multi-agent","features":["graph-based workflows","sequential, concurrent, handoff, and group patterns","A2A + MCP","durable, restartable runs","human-in-the-loop","Python + .NET + Go"],"url":"https://learn.microsoft.com/en-us/agent-framework/","github":"https://github.com/microsoft/agent-framework","notes":"Microsoft's successor to both AutoGen and Semantic Kernel, each of which ships a migration guide to it. Public preview landed October 1, 2025 and 1.0 shipped for Python and .NET on April 2, 2026, with a Go SDK in a separate repo. The default Microsoft pick for new multi-agent projects; weekly figure is PyPI agent-framework only and excludes NuGet."},{"id":"autogen","name":"AutoGen","vendor":"Microsoft","languages":["python"],"version":"0.7.x (final)","released":"2023-09-25","license":"CC-BY-4.0","githubStarsK":61.5,"weeklyInstallsK":86,"category":"multi-agent","features":["conversable agents","group chat","code execution","human approval","event-driven"],"url":"https://microsoft.github.io/autogen/","github":"https://github.com/microsoft/autogen","notes":"Microsoft Research multi-agent framework, now in maintenance mode: no new features, community managed, and limited to bug fixes and security patches. The last release was python-v0.7.5 on September 30, 2025. Microsoft points new users to Microsoft Agent Framework; listed for its large existing user base."},{"id":"crewai","name":"CrewAI","vendor":"crewAI Inc.","languages":["python"],"version":"1.15.x","released":"2023-12-12","license":"MIT","githubStarsK":59.9,"weeklyInstallsK":669,"category":"multi-agent","features":["roles","tasks","crews","flows","tool registry"],"url":"https://crewai.com","github":"https://github.com/crewAIInc/crewAI","notes":"Role-based multi-agent framework. The \"Agent + Task + Crew\" abstraction maps cleanly onto how teams actually think about delegation. Hugely popular for marketing, research, and sales agent setups."},{"id":"pydantic-ai","name":"Pydantic AI","vendor":"Pydantic","languages":["python"],"version":"2.57.x","released":"2024-12-02","license":"MIT","githubStarsK":20.7,"weeklyInstallsK":1474,"category":"agent-orchestration","features":["type-safe agents","pydantic models","streaming","logfire integration","function calling","realtime voice"],"url":"https://pydantic.dev/pydantic-ai","github":"https://github.com/pydantic/pydantic-ai","notes":"Type-safe agent framework from the Pydantic team, now on its 2.x line and pitched as typed end to end across agents, realtime voice, image generation, and embeddings. Strong fit for Python codebases that already use Pydantic for data validation. Draws about 1.5 million weekly PyPI installs as of October 2026."},{"id":"mastra","name":"Mastra","vendor":"Mastra","languages":["typescript"],"version":"1.78.x","released":"2024-09-12","license":"Apache-2.0 (ee/ directories under a separate enterprise license)","githubStarsK":28.9,"weeklyInstallsK":1330,"category":"agent-orchestration","features":["workflows","agents","rag","evals","memory","voice"],"url":"https://mastra.ai","github":"https://github.com/mastra-ai/mastra","notes":"TypeScript-first agent framework, built by the team behind Gatsby. Targets Node.js and Next.js stacks where most agent libraries are Python-only. Strong DX with first-class evals. Licensed Apache-2.0, with enterprise features kept under ee/ directories. Weekly figure is npm @mastra/core."},{"id":"openai-agents-sdk","name":"OpenAI Agents SDK","vendor":"OpenAI","languages":["python","typescript"],"version":"0.24.x","released":"2025-03-12","license":"MIT","githubStarsK":30.1,"weeklyInstallsK":3232,"category":"sdk","features":["handoffs","guardrails","tracing","sessions","voice agents","sandbox agents"],"url":"https://openai.github.io/openai-agents-python/","github":"https://github.com/openai/openai-agents-python","notes":"OpenAI's official client-side agent SDK. Handoffs are first-class for multi-agent, and it integrates tightly with OpenAI tracing. For a fully hosted alternative, OpenAI's Agents API (public beta since September 10, 2026) runs the Codex harness server-side. Use the SDK if you want the loop in your own process; otherwise LangGraph or Pydantic AI are usually a better multi-provider fit."},{"id":"claude-agent-sdk","name":"Claude Agent SDK","vendor":"Anthropic","languages":["python","typescript"],"version":"0.2.x (Python), 0.3.x (TypeScript)","released":"2025-09-29","license":"MIT","githubStarsK":8.3,"weeklyInstallsK":7652,"category":"sdk","features":["claude code core","mcp servers","hooks","subagents","slash commands"],"url":"https://code.claude.com/docs/en/agent-sdk/overview","github":"https://github.com/anthropics/claude-agent-sdk-python","notes":"The Claude Code core, repackaged as a programmable SDK. Native MCP, hooks, and subagent orchestration. Best fit if your agent needs the same surface area as Claude Code. Weekly figure is PyPI only; the TypeScript package drew about 7.3 million npm downloads the same week."},{"id":"vercel-ai-sdk","name":"Vercel AI SDK","vendor":"Vercel","languages":["typescript","javascript"],"version":"7.x","released":"2023-04-21","license":"Apache-2.0","githubStarsK":27.4,"weeklyInstallsK":18353,"category":"sdk","features":["streaming UI","tool calling","structured output","multi-provider","react/svelte/vue hooks","agents"],"url":"https://ai-sdk.dev","github":"https://github.com/vercel/ai","notes":"TypeScript-first AI SDK, now on v7 with its docs at ai-sdk.dev. Best in class for streaming UI in Next.js and React apps, with agent loops and multi-step tool calling built in. The most-used AI library in JS-land by npm downloads."},{"id":"agno","name":"Agno","vendor":"Agno","languages":["python"],"version":"3.x","released":"2024-08-15","license":"Apache-2.0","githubStarsK":42.8,"weeklyInstallsK":393,"category":"agent-orchestration","features":["multi-modal agents","memory","knowledge bases","workflows","monitoring"],"url":"https://www.agno.com","github":"https://github.com/agno-agi/agno","notes":"Phidata rebranded, now on its 3.x line under Apache-2.0 and positioned as a way to build, run, and manage agent platforms. Lightweight Python framework with first-class multimodal (image, audio, video) agents and strong observability through Agno Platform."},{"id":"smolagents","name":"smolagents","vendor":"Hugging Face","languages":["python"],"version":"1.26.x","released":"2024-12-31","license":"Apache-2.0","githubStarsK":29.9,"weeklyInstallsK":108,"category":"agent-orchestration","features":["code agents","hub integration","multi-step","minimal abstractions"],"url":"https://huggingface.co/docs/smolagents","github":"https://github.com/huggingface/smolagents","notes":"Minimal agent framework from Hugging Face. CodeAgent expresses tool use as Python code instead of JSON, which often produces better trajectories on agentic benchmarks. Releases have slowed (1.26.0 shipped May 29, 2026), though the repo is still maintained."},{"id":"haystack","name":"Haystack","vendor":"deepset","languages":["python"],"version":"3.4.x","released":"2020-04-13","license":"Apache-2.0","githubStarsK":26.8,"weeklyInstallsK":108,"category":"rag","features":["pipelines","document stores","agents","evaluation","hybrid retrieval"],"url":"https://haystack.deepset.ai","github":"https://github.com/deepset-ai/haystack","notes":"One of the original RAG frameworks (predates the LLM era), now on the 3.x line. Modular pipelines and agent workflows with explicit control over retrieval, routing, and memory. Strong for enterprise RAG and competitive with LangGraph for retrieval-heavy workflows."},{"id":"browser-use","name":"browser-use","vendor":"Browser Use","languages":["python"],"version":"0.13.x","released":"2024-11-15","license":"MIT","githubStarsK":118.5,"weeklyInstallsK":1618,"category":"browser-agent","features":["playwright","vision","click/type/scroll","multi-tab","file download"],"url":"https://browser-use.com","github":"https://github.com/browser-use/browser-use","notes":"Open-source browser automation framework for AI agents. Passes pages to LLMs as DOM plus screenshots so the model can decide each action. Past 118,000 GitHub stars as of October 2026."},{"id":"google-adk","name":"Agent Development Kit (ADK)","vendor":"Google","languages":["python","typescript","multi"],"version":"2.12.x","released":"2025-04-09","license":"Apache-2.0","githubStarsK":21.9,"weeklyInstallsK":2288,"category":"agent-orchestration","features":["code-first agents","graph workflows","multi-agent","evaluation","deployment tooling","Python + TypeScript + Go + Java + Kotlin"],"url":"https://adk.dev","github":"https://github.com/google/adk-python","notes":"Google's open-source, code-first framework for building, evaluating, and deploying agents. Optimized for Gemini but model-agnostic and deployment-agnostic. ADK 2.0 brought graph workflows, and SDKs now span Python, TypeScript, Go, Java, and Kotlin. Weekly figure is PyPI google-adk only."},{"id":"pipecat","name":"Pipecat","vendor":"Daily","languages":["python"],"version":"1.13.x","released":"2024-04-01","license":"BSD-2-Clause","githubStarsK":16.6,"weeklyInstallsK":324,"category":"voice-agent","features":["STT","TTS","WebRTC","turn detection","multi-modal"],"url":"https://pipecat.ai","github":"https://github.com/pipecat-ai/pipecat","notes":"Real-time voice agent framework maintained by Daily and the community, now past 1.0. Pipelines connect STT (Deepgram, Whisper), LLMs, and TTS (Cartesia, ElevenLabs) over WebRTC with sub-second turn-taking. The default OSS stack for production voice agents."}]}