Long-Term Context

Persistent Memory

Agents that remember. Store facts, preferences, and context across conversations — per agent or shared across your entire system.

Live Demo

Watch memory in action

As the conversation flows, the agent stores facts and recalls them later to give personalized, context-aware responses.

Conversation
Conversation will appear here...
Memory Store
user_name
company
tech_stack
Write — storing new info
Read — recalling for context
Recalled — used in response

Memory Types

Two layers of memory

Agent Memory

Private to each agent. Store facts, learned patterns, and preferences that only this agent needs. Perfect for specialization.

Per-agent isolation
Session persistence
Key-value storage
Automatic context injection

Shared Memory

Accessible by all agents in the system. Share knowledge, coordinate state, and build collective understanding across your agent team.

Cross-agent access
Global knowledge base
Conflict-free reads/writes
Orchestrator-level coordination

Storage

Pluggable backends

Choose where memory lives. Start simple with in-memory or file storage, then scale to any backend you need.

Built-in

In-Memory

Fast, ephemeral storage for development and testing. Data lives in Python dicts.

Built-in

File System

Persist memory to JSON files on disk. Survives restarts with zero setup.

Built-in

SQLite

Lightweight relational storage. Great for structured queries and local production use.

Extensible

Custom

Implement the CacheStore interface for Redis, PostgreSQL, MongoDB, or any storage you need.

Architecture

Context management

Three layers work together to give agents the right context at the right time, without exceeding token limits.

Active ContextRecent messages in windowSummariesCompressed older contextPersistent StoreLong-term memory on diskAgentGets the right contextFast accessOn demandWhen needed

Agents that remember

Give your agents context that persists. No more starting from scratch.