plastic-labs/honcho

Overview

This skill integrates Honcho memory and social cognition into Python or TypeScript codebases to build stateful, personalized AI agents. It guides setup of the SDK, peer and session modeling, message wiring, and implementing the dialectic chat endpoint so agents can query user context on-demand. Use it to add long-term user modeling, contextual prompts, or tool-based context retrieval for agent frameworks and bots.

How this skill works

The integration creates Honcho clients (sync or async) and instantiates peers for every participant (users, assistants, bots). Sessions group conversations and accept messages that Honcho reasons over asynchronously to build representations. Your agent accesses memory via peer.chat(query) or session.context(), or by exposing Honcho as a tool the agent can call during reasoning.

When to use it

• Adding Honcho SDK to a Python or TypeScript project (FastAPI, Express, Next.js, Flask, Django).
• Wiring message flows so every conversational exchange is stored and modeled.
• Implementing dialectic chat as a tool for agents using function calling or tool APIs.
• Prefetching targeted user attributes to include in LLM prompts.
• Integrating Honcho with bot frameworks (nanobot, openclaw, picoclaw) that have session and peer concepts.

Best practices

• Use a single workspace per product and keep workspace_id consistent across environments.
• Create peers for all entities; set AI peers to observe_me=False to avoid modeling the agent.
• Match the Honcho client (sync vs async) to your framework handlers (def vs async def).
• Persist every message exchange to sessions so Honcho can build representations over time.
• Prefer dialectic chat as a tool call for agents; fallback to pre-fetch when tool calling is not available.

Example use cases

• Expose Honcho as an OpenAI function/tool so an agent queries user goals on-demand during reasoning.
• Prefetch user attributes (communication style, expertise, goals, preferences) and inject them into system prompts.
• Add peers and sessions to a FastAPI chat endpoint and store messages after each send/receive cycle.
• Integrate with a bot framework (nanobot/picoclaw) by creating peers for each bot and the user, and adding session messages.
• Build personalization features that answer questions like 'What does this user care about?' using peer.chat.