buildagentic

Tracker

Which agent frameworks are people actually using?

GitHub stars are a weak proxy for adoption and the only one every project publishes. Ranked lowest to highest, coloured by what each framework is for, so the shape of the field is visible before any single name is.
Every dot is a row in Agent engineering frameworks, with the source that reported it and the moment it was read.18 tracked →
GitHub stars
30 days of readings

LangChain leads on GitHub stars with 147k, 5.1x the median of 29k.

HighestLangChain · 147kMedianSemantic Kernel · 29kLowestGuardrails · 7kTracked18

Dot height = GitHub stars, dot colour = what the framework is for. Left to right is rank, lowest first — the horizontal position carries no other meaning.

  • orchestration6
  • multi-agent3
  • evaluation2
  • rag2
  • structured-output2
  • memory1
  • optimisation1
  • validation1

memory, optimisation, validation share the neutral swatch — the palette carries six distinct colours. Point at a mark to name it.

037k73k110k147kGitHub starsLowestHighestGuardrails7kSemantic Kernel29kLangChain147kGuardrails7kvalidationSynced from source, 23h agoInstructor14kstructured-outputSynced from source, 23h agoRagas16kevaluationSynced from source, 23h agoDeepEval18kevaluationSynced from source, 23h agoPydantic AI20kstructured-outputSynced from source, 23h agoLetta25kmemorySynced from source, 23h agoHaystack27kragSynced from source, 23h agoVercel AI SDK27korchestrationSynced from source, 23h agoMastra28korchestrationSynced from source, 23h agoSemantic Kernel29korchestrationSynced from source, 23h agosmolagents29kmulti-agentSynced from source, 23h agoDSPy38koptimisationSynced from source, 23h agoLangGraph42korchestrationSynced from source, 23h agoAgno42korchestrationSynced from source, 23h agoLlamaIndex52kragSynced from source, 23h agoCrewAI59kmulti-agentSynced from source, 23h agoAutoGen61kmulti-agentSynced from source, 23h agoLangChain147korchestrationSynced from source, 23h ago
Agent engineering frameworks →

History

What has moved since we started watching

30 distinct days of readings so far. Every observation is kept; none is overwritten.

  • LangChain

    147k +2k over 30d

    LangChain: 37 readings, 145k to 147k.

  • LangGraph

    42k +2k over 30d

    LangGraph: 37 readings, 40k to 42k.

  • CrewAI

    59k +1k over 30d

    CrewAI: 37 readings, 58k to 59k.

  • Mastra

    28k +833 over 30d

    Mastra: 37 readings, 27k to 28k.

  • DSPy

    38k +648 over 30d

    DSPy: 37 readings, 38k to 38k.

  • Pydantic AI

    20k +643 over 30d

    Pydantic AI: 36 readings, 19k to 20k.

  • DeepEval

    18k +576 over 30d

    DeepEval: 36 readings, 18k to 18k.

  • AutoGen

    61k +510 over 30d

    AutoGen: 37 readings, 61k to 61k.

  • Vercel AI SDK

    27k +509 over 30d

    Vercel AI SDK: 37 readings, 26k to 27k.

  • smolagents

    29k +477 over 30d

    smolagents: 36 readings, 29k to 29k.

  • LlamaIndex

    52k +439 over 30d

    LlamaIndex: 37 readings, 52k to 52k.

  • Letta

    25k +436 over 30d

    Letta: 36 readings, 24k to 25k.

  • Agno

    42k +423 over 30d

    Agno: 36 readings, 42k to 42k.

  • Ragas

    16k +374 over 30d

    Ragas: 35 readings, 15k to 16k.

  • Haystack

    27k +273 over 30d

    Haystack: 37 readings, 26k to 27k.

  • Instructor

    14k +160 over 30d

    Instructor: 36 readings, 14k to 14k.

  • Guardrails

    7k +127 over 30d

    Guardrails: 37 readings, 7k to 7k.

  • Semantic Kernel

    29k +103 over 30d

    Semantic Kernel: 37 readings, 28k to 29k.

More trackers

Other questions