Honcho

Honcho is an open-source memory library with a managed service for building stateful agents. It works with any model, framework, or architecture and enables agents to build and maintain state about entities—users, agents, groups, ideas, and more. As a continual learning system, it understands entities that change over time and aims to help agents retain context across interactions.

The system uses four storage primitives—Workspaces, Peers, Sessions, and Messages—to store data and trigger reasoning. Messages are written to Honcho, stored and indexed, and then reasoned over by Neuromancer-based models to produce conclusions about each peer that can be queried for richer context.

Honcho emphasizes token efficiency and scalable context management. It provides granular control over retrieval with search, token budgets, summaries, and peer scoping. It also exposes formats to convert context for popular LLMs, such as OpenAI and Anthropic, to integrate with existing workflows.

Use cases

• Maintain state across sessions for users, agents, and groups
• Reduce token usage by retaining only relevant context
• Integrate memory with common AI toolchains