> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orxhestra.com/llms.txt
> Use this file to discover all available pages before exploring further.
# orxhestra
> Open-source Python framework for orchestrating multi-agent AI systems. Async streaming, composable agent hierarchies, planners, A2A & MCP integration.
Multi-agent orchestration framework for Python — **compose entire agent teams in YAML**, run them in the terminal, or expose them as an A2A server. Python SDK below when you need full programmatic control.
Build and run your first agent in under 5 minutes
Declarative multi-agent teams — no Python wiring needed
Source code, examples, and the plan file
```bash theme={null}
pip install 'orxhestra[cli,anthropic]'
orx # interactive REPL with the built-in coding agent
orx my-agents.yaml # run your own YAML spec
orx my-agents.yaml --serve -p 9000 # same spec as an A2A server
```
***
## Features
Declarative multi-agent teams — LLM agents, loops, pipelines, fan-out, transfer routing — in a single file. Built-in registries for custom agent types, LLM providers, built-in tools, and whole new tool kinds.
Every agent is an async generator yielding typed events — thoughts, tool calls, results, streaming chunks. Built for real-time UIs.
Nest agents freely. Chain sequentially, run in parallel, loop until done, or wrap as tools. Plain Python asyncio.
Per-turn planning hooks inject dynamic instructions and post-process responses. Built-in PlanReAct and TaskPlanner.
Pluggable session service persists events and state deltas automatically. In-memory included, bring your own database.
Function tools, agent-as-tool, transfer handoffs, exit loop, long-running tools, tool confirmation, and a `register_tool_resolver` hook for whole new tool kinds — all built in.
Automatic LLM-based summarization of old session events. Keep context windows manageable across long conversations.
Opt-in Ed25519 event signing, DID-based agent identity, hash-chained audit logs, and pluggable `AttestationProvider`s — all wireable from a YAML block.
First-class tracing via AgentConfig callbacks. Works with Langfuse, LangSmith, or any LangChain callback handler.
***
## Works with any LangChain LLM
```bash theme={null}
pip install orxhestra[openai]
```
```bash theme={null}
pip install orxhestra[anthropic]
```
```bash theme={null}
pip install orxhestra[google]
```
```bash theme={null}
pip install orxhestra
# + your preferred langchain-* provider package
```