Cookbook¶

Practical recipes demonstrating common patterns with OpenAgentMesh. Each recipe is a self-contained code sample you can copy and run.

To run any recipe against a local embedded mesh:

import asyncio
from openagentmesh import AgentMesh

async def run():
    async with AgentMesh.local() as mesh:
        # paste the recipe's main() body here
        ...

asyncio.run(run())

Or against your own mesh started with oam mesh up:

async with AgentMesh() as mesh:     # connects to nats://localhost:4222
    ...

Available Recipes¶

Recipe	Pattern
Multi-Process Agents	Provider/consumer on a shared bus
Multi-Module Projects	Scaling beyond a single file
Shared Plan Coordination	CAS-based concurrent state
LLM-Driven Tool Selection	Runtime discovery for LLM tools
MCP Bridge	Serve mesh agents to MCP clients
Secured Mesh	JWT credentials and roles
Error Handling	Retry, fallback, monitoring
Surviving Agent Failures	Death notices and fast failover
Observing the Mesh	Live log tailing and level control
Tracking LLM Usage	Per-agent token and cost attribution
Admin UI	Browse, invoke, and watch from the browser
Lifecycle-Gated Agents	Condition-driven agent on/off
Automatic Load Balancing	Queue group scaling
Reactive Pipeline	KV watch coordination
Parallel RAG Indexing	ObjectStore + queue groups

Interactive demos

For an interactive experience, try oam demo. It starts a local mesh with sample agents you can explore from another terminal.