edit on github↗

Documentation

Agentspan is a durable runtime for AI agents. Your code runs in your process. Execution state lives on the server.

Agentspan is a durable runtime for AI agents. Execution state lives server-side, so crashes, restarts, and deployments do not lose work. Write agents natively or wrap an existing LangGraph, OpenAI Agents SDK, or Google ADK agent in one line.

Getting Started

  • Why Agentspan - Why agents fail in production, and how Agentspan solves it.
  • Quickstart - Build your first agent in 5 minutes.

Concepts

  • Agents - The Agent class, parameters, results, and handles.
  • Tools - @tool, http_tool(), api_tool(), mcp_tool(), credentials, and approval-required tools.
  • Multi-Agent Strategies - Sequential, parallel, handoff, router, and nested agent coordination.
  • Guardrails - Input and output safety, retry, block, and fix behavior.
  • Memory - Conversation history and semantic search across sessions.
  • Streaming - Runtime events, async execution, and HITL with streams.
  • Testing - mock_run, expect, record/replay, pytest, and evaluation helpers.

Deployment

Examples

Reference

Next Why Agentspan →