State Management

State in LeafMesh is automatic and stored in Redis. This page covers patterns for working with state within agent executions.

Automatic State

LeafMesh auto-stores these without explicit calls:

Session metadata and conversation history
Agent yields after each execution
Mesh communications between agents
Coordination decisions and observations

The store behind all of this is configured in .env:

# .env
LEAFMESH_AGENCY_NAME=my-agency
LEAFMESH_STORE=redis
REDIS_HOST=localhost
REDIS_PORT=6379
# 1 hour for general data
REDIS_DEFAULT_TTL=3600
# 7 days for sessions (the default)
REDIS_SESSION_TTL=604800

From 2.4.482 the store is set only in .env, never in the YAML.

Working with State in Intelligence Functions

Use the context and input_data parameters to access state:

async def counter_agent(llm_response, input_data, context):
    session_id = context.get("session_id", "default")
    yields = context.get("yields", {})

    return {**llm_response, "session": session_id}

Accessing Upstream Yields

When agent B is called by agent A via can_call, agent A's yields are passed as input_data:

async def checker(llm_response, input_data, context):
    # input_data contains the upstream agent's yields
    claimed_answer = input_data.get("answer")
    return {"is_correct": claimed_answer == 42}

Session-Scoped Storage

All state in LeafMesh is automatically scoped per session — session metadata, conversation history, agent yields, and mesh communications are kept isolated so different sessions never collide.

These records are managed automatically. Browse them in Studio's Sessions tab, or query them via the platform's REST API.

Next Steps

Short-Term Memory — Full session management
Redis Integration — Redis configuration
Context Threading — Parallel session threads

LeafMesh — Explicit, inspectable state