Agent Data Store

The Agent Data Store provides persistent key-value storage for agents, enabling them to save and retrieve data across executions.

Overview

The data store enables:

Persistence: Save results between agent runs
Caching: Avoid redundant expensive operations
Cross-agent communication: Share data between agents
Incremental processing: Build up results over multiple executions

Data is automatically scoped per-user, so agents can only access their owner's data.

Quick Example

The data_store object is automatically available in every agent:

async def run(input_dict: dict, tools: dict) -> dict:
    # Check cache
    cached = data_store.get("analysis-result")
    if cached:
        return {"result": cached, "source": "cache"}
    
    # Compute and cache
    result = await expensive_analysis()
    data_store.set("analysis-result", result)
    
    return {"result": result, "source": "computed"}

Documentation

Detailed documentation is available in the data-store directory:

Quick Start - Get started in 5 minutes
API Reference - Complete method documentation
Schema - Document structure
Namespaces - Organizing data
Patterns - Common usage recipes
Cross-Agent Workflows - Multi-agent data sharing
Troubleshooting - Common issues
Administration - Database management

Key Features

Namespaces

Organize data into logical groups:

# Different namespaces for different purposes
files = data_store.use_namespace(f"files:{repo}")
summaries = data_store.use_namespace(f"summary:{repo}")
cache = data_store.use_namespace("api-cache")

Namespace Discovery

Find what data exists:

namespaces = data_store.list_namespaces()
# Returns: ["default", "files:my-repo", "cache:github", ...]

Batch Operations

Efficient bulk operations:

# Set multiple values
data_store.set_many({"key1": val1, "key2": val2})

# Get multiple values
results = data_store.get_many(["key1", "key2"])