Context is precious. Execution is sacred. Trust is earned.Python 3.10–3.13

Universal Tool Orchestration Platform

Give every agent the right tools at the right moment.

Agent-Gantry is a Python 0.8.0 library for semantic tool routing, secure execution, provider schema conversion, MCP/A2A interoperability, and framework adapters. It replaces prompt-stuffing hundreds of tools with a governed retrieval-and-execution layer.

Start the journey → API reference Microsoft Agent Framework →

~79–90%

Less tool context

Retrieve top-k tools instead of injecting the entire registry into every model call.

1×

Register once

Transcode to OpenAI, Anthropic, Gemini, framework-native adapters, MCP, and A2A paths.

0-trust

Execution guardrails

Policies, capabilities, timeouts, retries, rate limits, circuit breakers, and telemetry.

Interactive implementation map

Register

Decorate ordinary Python functions and let Gantry infer schemas from type hints and docstrings.

@gantry.register(tags=["finance"])
def calculate_tax(amount: float) -> float:
    """Calculate tax."""
    return amount * 0.08

Current integration surface

LLM SDKs

  • OpenAI / Azure / OpenRouter
  • Anthropic Claude
  • Google GenAI
  • Google Vertex AI
  • Groq
  • Mistral via OpenAI-compatible endpoint

Frameworks

  • Microsoft Agent Framework
  • LangChain
  • LangGraph
  • LlamaIndex
  • CrewAI
  • AutoGen
  • Semantic Kernel
  • Google ADK
  • Pydantic AI
  • OpenAI Agents SDK
  • Smolagents
  • Haystack
  • Agno
  • Strands Agents
  • DSPy

Protocols and storage

  • MCP server/client routing
  • A2A server/executor
  • LanceDB persistence
  • Qdrant / Chroma / pgvector adapters
  • Nomic / OpenAI / sentence-transformers embeddings
  • Cohere / cross-encoder rerankers