Inno Agent

// Overview

Inno Agent is a single-learner study companion built on the Pi coding-agent SDK without modifying the SDK's kernel — every learning behaviour is added through registered tools and one extension hook. Its premise is that education is a different target from general-purpose coding agents: what matters is personalised explanation, misconception diagnosis, exercise generation, feedback, review scheduling, privacy and low-latency continuous interaction.

Long-term support is organised into three explicit memory layers with different lifecycles. The L1 learner profile holds goals, knowledge states, misconceptions and preferences, summarised into a context pack injected every turn; the learner can inspect and correct it, and unevidenced labels are forbidden. The L2 layer is a native wiki of human-readable pages with hybrid retrieval (BM25 plus a knowledge graph) and ingestion of PDF, Office and image files. The L3 layer indexes past sessions into SQLite FTS5 for threshold-gated recall across conversations. Facts that should shape future teaching are written to these layers through tools rather than left in chat replies.

Around the memory sits a learning loop: a cron scheduler whose jobs you create in natural language, personal IM channels (Feishu natively, WeChat via iLink QR login or a bridge) for reminders, and a Practice Lab — a workspace-scoped in-browser terminal whose run records the agent can read. It ships as an Electron desktop app, a React web UI served by a Node server, and an `inno` terminal CLI, all sharing the same runtime state. It is deliberately personal: one process, one active session, no multi-user auth — a team runs one instance per person. Models come from any OpenAI-completions or Anthropic-messages endpoint, including local ones through Ollama. The project is MIT-licensed and accompanied by a technical report.

// Getting started

Inno Agent runs as a desktop app, a web server or a terminal CLI. The commands below come from the repository README; the source install needs Node.js 20.6 or newer.

~/.inno-agent/config/config.json

git clone https://github.com/hhyqhh/inno-agent.git
cd inno-agent

npm install
npm run build

mkdir -p runtime/config runtime/data runtime/skills workspace
cp config.example.json runtime/config/config.json

{
  "defaultProvider": "innospark",
  "defaultModel": "claude-sonnet-4-6",
  "providers": {
    "innospark": {
      "baseUrl": "https://api.example.com",
      "api": "anthropic-messages",
      "apiKey": "replace-me",
      "models": [{ "id": "claude-sonnet-4-6", "name": "Claude Sonnet 4.6" }]
    }
  },
  "memory": { "l1Enabled": true, "l2Enabled": true, "l3Enabled": true }
}

npm run server -- --home ./runtime --workspace ./workspace --port 3000

npm run start           # CLI (terminal agent, no HTTP)
npm run electron        # desktop app locally
npm run server:sandbox  # web UI with OS-level sandbox (requires ripgrep)
docker compose up -d    # Docker

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Reach for it when you want a tutor that keeps an inspectable, correctable model of what you know and where you go wrong
• Reach for it when study notes should accumulate into a searchable wiki the agent can query in later sessions
• Reach for it when review reminders should arrive on Feishu or WeChat on a schedule instead of waiting for you to open a chat
• Reach for it when teachers need one-click presets for lesson plans, classroom exercises or slide decks without configuring an agent

// How Inno Agent compares

Inno Agent alongside other open-source learning & tutoring tools AI/TLDR tracks, ranked by GitHub stars.