docs_skill

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 storage, and recommended patterns for querying user context. Use it to add continual learning, representations, and context-aware responses to existing chat or agent systems.

How this skill works

The skill inspects your repo to detect language, frameworks, existing LLM integrations, session handling, and message flow. After a short interview to map peers, integration pattern, and session logic, it provides concrete implementation code: initializing the Honcho client, creating peers, configuring sessions, adding messages, and wiring the dialectic chat endpoint as a tool or pre-fetch step. It also includes verification and checklist steps to ensure messages and workspace configuration are correct.

When to use it

• Adding long-term user memory and personalization to chatbots or agents
• Integrating Honcho SDK into Python (FastAPI, LangChain) or TypeScript (Express, Next.js) projects
• Implementing function-calling or tool-call patterns for agent access to user context
• Mapping existing conversation/session models to Honcho sessions
• Ensuring consistent message storage and workspace configuration across the app

Best practices

• Use a single workspace ID for the entire application to centralize memory
• Create peers for all entities (human users and AI assistants); set AI peers observe_me=false
• Choose the integration pattern that fits your architecture: tool-call for agents, pre-fetch for simpler flows, or context() for LLM-ready history
• Always call add_messages() after exchanges so Honcho can build user representations
• Document and surface HONCHO_API_KEY and workspace ID in environment configuration

Example use cases

• Expose Honcho as an OpenAI function/tool so an agent queries user context on-demand during reasoning
• Pre-fetch targeted user attributes (communication style, goals, preferences) to build a system prompt before each LLM call
• Use session.context() to produce LLM-formatted conversation history with summaries for long threads
• Add multi-peer sessions for group chats with per-peer observation settings
• Stream Honcho chat responses for real-time UX while feeding exchanges back into Honcho