hhyqhh/inno-agent

An open-source personal learning agent with three-layer memory (learner profile / wiki knowledge base / cross-conversation recall), a proactive scheduler, personal IM channels, and a workspace-scoped Practice Lab — built on the Pi SDK.

What it solves

Inno Agent addresses the limitations of general-purpose coding agents when applied to education. It focuses on personalized learning support—such as misconception diagnosis, exercise generation, and review scheduling—rather than open-ended software engineering. It prevents the "flat chat" problem by separating learner state, archived knowledge, and dialogue history into distinct memory layers.

How it works

Built on the Pi coding-agent SDK, the system implements a learning loop through four layers:

  1. User Interfaces: Accessible via a Desktop app (Electron), Web UI (React), Terminal CLI, and IM channels (Feishu/WeChat).
  2. Application Layer: Manages channel adapters, a cron-based proactive scheduler, and a "Practice Lab" with an in-browser terminal.
  3. Agent Runtime: Uses the Pi SDK to orchestrate LLM providers (OpenAI, Anthropic, DeepSeek, Ollama) and registered tools.
  4. Layered Memory:
    • L1 (Learner Profile): Tracks goals, knowledge states, and preferences.
    • L2 (Native Wiki): A human-readable knowledge base using hybrid retrieval (BM25 + knowledge graph).
    • L3 (Session Recall): SQLite-indexed history for cross-conversation retrieval.

Who it’s for

Individual learners who want a personalized AI companion to manage their study process, track their progress, and practice skills in a sandboxed environment.

Highlights

  • Three-Layer Memory: Separates learner profiles, a structured wiki, and session history to ensure personalization is evidence-driven.
  • Proactive Scheduling: Allows the agent to create and run cron jobs in natural language for reminders and reviews.
  • Practice Lab: Includes a workspace-scoped web terminal for hands-on practice with records the agent can analyze.
  • Multi-Channel Access: Integrates with Feishu and WeChat for seamless interaction.
  • Session Replay: A standalone showcase site that can replay real recorded sessions using the actual product UI.

Related

  • Project
    vellum-ai/vellum-assistant

    A personal AI assistant with a sophisticated eight-type memory system that learns user behavior and operates across multiple apps and devices.

  • Project
    zhayujie/CowAgent

    An open-source AI assistant framework that proactively plans tasks, manages long-term memory and knowledge, and controls computers across multiple messaging platforms.

  • Project
    fuyuxiang/echo-agent

    A self-hosted, long-running AI agent featuring a four-layer cognitive memory system and a self-evolution engine to improve skills over time.

  • Project
    beilusaiying/always-accompany

    A multi-purpose AI agent framework focused on context and attention mechanisms, providing layered memory and dynamic retrieval for long-term companionship, coding, and work.

  • Project
    HKUDS/nanobot

    An ultra-lightweight, self-hosted personal AI agent framework that combines long-term memory, MCP integrations, and scheduled automation across WebUI, terminal, and various chat apps.