The agentic AI leaderboard▌
Open-source agent frameworks and tools, ranked live by GitHub stars - with license, pricing and activity. Adoption signal, not a verdict.
Agentic frameworks, ranked live
| # | framework | category | lang | license | pricing | stars | forks | last push |
|---|---|---|---|---|---|---|---|---|
| 01 | n8nFair-code workflow automation with native AI agent nodes - connect anything, then let agents act across your stack.github.com/n8n-io/n8n -> | workflow automation | TypeScript | Sustainable Use License | source available | 202.8k | 60.5k | 2026-08-29 |
| 02 | LangflowA visual framework for multi-agent and RAG apps, built on top of LangChain.github.com/langflow-ai/langflow -> | visual builder | Python | MIT | open source | 153.9k | 10.0k | 2026-08-29 |
| 03 | DifyAn open-core platform for building LLM apps and agent workflows visually, then deploying them to production.github.com/langgenius/dify -> | platform | TypeScript | Dify Open Source License | open core | 153.8k | 24.3k | 2026-08-29 |
| 04 | AutoGenMicrosoft's framework for multi-agent conversation systems, from agent chat to autonomous event-driven workflows.github.com/microsoft/autogen -> | orchestration | Python | CC-BY-4.0 | open source | 60.7k | 9.2k | 2026-04-15 |
| 05 | CrewAIRole-based multi-agent orchestration. Define crews of agents with distinct roles, goals and tools, then run them as a team.github.com/crewAIInc/crewAI -> | orchestration | Python | MIT | open source | 57.8k | 8.3k | 2026-08-28 |
| 06 | FlowiseA low-code builder for LLM apps and agent flows. Drag, drop, deploy.github.com/FlowiseAI/Flowise -> | visual builder | TypeScript | Apache-2.0 | open core | 55.4k | 25.0k | 2026-08-13 |
| 07 | LlamaIndexThe data framework for LLM apps - connectors, indexing and retrieval that ground agents in your own data.github.com/run-llama/llama_index -> | data / rag | Python | MIT | open source | 51.9k | 8.1k | 2026-08-29 |
| 08 | LangGraphStateful orchestration for multi-agent apps, from the LangChain team. Graphs of agents and tools with built-in memory and human-in-the-loop.github.com/langchain-ai/langgraph -> | orchestration | Python | MIT | open source | 40.7k | 6.9k | 2026-08-28 |
| 09 | OpenAI Agents SDKA lightweight framework from OpenAI for multi-agent workflows built around models, tools and handoffs.github.com/openai/openai-agents-python -> | agent sdk | Python | MIT | open source | 29.1k | 4.6k | 2026-08-28 |
| 10 | SmolagentsA minimal code-agent library from Hugging Face. Agents that write and run their own Python, in a few lines.github.com/huggingface/smolagents -> | agent sdk | Python | Apache-2.0 | open source | 29.0k | 2.9k | 2026-08-25 |
| 11 | Semantic KernelEnterprise AI orchestration SDK. Plug models into an existing C#, Python or Java codebase with agents, plugins and memory.github.com/microsoft/semantic-kernel -> | orchestration | C# | MIT | open source | 28.5k | 4.7k | 2026-08-27 |
| 12 | HaystackProduction RAG and agent framework from deepset. Composable pipelines for search, retrieval and agentic workflows.github.com/deepset-ai/haystack -> | data / rag | Python | Apache-2.0 | open source | 26.4k | 3.0k | 2026-08-29 |
| 13 | Pydantic AIType-safe agents from the Pydantic team. Structure inputs and outputs, get validated results and type-checked tool calls.github.com/pydantic/pydantic-ai -> | agent sdk | Python | MIT | open source | 19.6k | 2.6k | 2026-08-29 |
| 14 | Claude Agent SDKAnthropic's SDK for building Claude-powered agents with tools, subagents and a persistent session.github.com/anthropics/claude-agent-sdk-python -> | agent sdk | Python | MIT | open source | 8.0k | 1.3k | 2026-08-28 |
| 15 | WorkbenchBuildHaus's open-source framework for agentic software development - the structure, guardrails and workflow behind our client delivery.github.com/buildhaus/workbench -> | dev framework | TypeScript | MIT | open source | 13 | 2 | 2026-07-27 |
# ranked by github stars, refreshed live on page load. last synced: 2026-08-29.
How to read it
- stars
- An adoption signal from the public GitHub API - the number of people who have starred the repository. It tracks attention and community momentum, not quality or production-readiness.
- forks
- The number of forks, a rough proxy for derivative work and experimentation in the ecosystem.
- last push
- The most recent push to the default branch. An activity signal - stale dates can indicate a project that has slowed or been archived.
- license / pricing
- Recorded from each repository and reviewed by BuildHaus. Open source means a standard open license (MIT, Apache-2.0, CC-BY-4.0); open core and source available carry commercial restrictions.
# stars are a popularity metric, not a buying guide.
# a low rank is not a bad tool - it may be newer, narrower, or simply less famous than a big platform.
# Workbench is here because it is ours - and because we will not hide an honest number. It is the framework behind our client delivery, and it is early.
# the right framework is the one that fits your stack, team and requirements. If you are weighing frameworks, we will tell you straight what is worth using - and what is not.
Asked often
What does this leaderboard measure?
Adoption, not quality. Each entry is ranked by GitHub stars - a popularity signal that tracks attention and community, not production-readiness. Stars, forks and last-push dates are pulled live from the public GitHub API; the metadata (license, pricing, category) is reviewed by BuildHaus.
Is a higher rank a better tool?
No. Stars measure attention, not fit. The right framework depends on your use case, existing stack and requirements - the leaderboard is a map of the landscape, not a verdict on any single tool.
How is the data sourced?
Live metrics come from the public GitHub REST API and refresh on every page load (cached in your browser for 60 minutes to be a good API citizen). License and pricing are recorded from each project's repository and reviewed by BuildHaus.
How often does it update?
The numbers refresh live whenever the page is loaded. The committed snapshot (shown before JavaScript runs, and to crawlers) is regenerated by a script that re-pulls the GitHub API.
$ book --type discovery --duration 30m --cost $0
Thirty minutes.
Zero dollars. Real answers.
Bring the problem, the half-built prototype, or just the hunch that agents could change how you work. We’ll tell you straight what’s worth building — and what isn’t.
Book a free 30-min discovery call-># no deck, no pressure, no obligation. exit code 0 either way.