# Inno Agent

> Open-source personal learning agent with three-layer memory, a proactive scheduler, IM channels, and a Practice Lab.

Inno Agent is an open-source personal learning agent published by hhyqhh under the MIT License. It organizes a learner profile, a native wiki knowledge base, and session records into a long-term learning loop, and is built on the Pi coding-agent SDK without modifying its kernel. It ships as a desktop app, a Web UI, and a terminal CLI that share the same runtime state.

## What It Is

Inno Agent is a single-learner companion for education rather than a general-purpose coding agent. It targets personalized explanation, misconception diagnosis, exercise generation, feedback, and review scheduling. Learning data is stored in local files by default, and models connect through configurable providers.

## How the layered memory works

The project describes three explicit memory layers:

- **L1 learner profile**: goals, knowledge states, misconceptions, and preferences, which the learner can inspect and correct.
- **L2 native wiki**: pages for sources, entities, concepts, and analyses, with hybrid retrieval (BM25 plus a knowledge graph) and PDF, Office, and image ingestion.
- **L3 session recall**: session history indexed in SQLite FTS5 with cross-conversation retrieval.

## Learning loop and tools

A cron scheduler runs jobs created in natural language, such as daily reviews and spaced repetition, and can push reminders through Feishu and WeChat channels. The Practice Lab is a workspace-scoped web terminal where the agent generates exercise files and suggested commands, the user runs them, and saved run records feed the next explanation. Skills and preset workspaces can be imported from a content hub, including lesson-plan, PPT, and VeryMath math-research templates.

## Setup and deployment

Users can download a macOS (Apple Silicon) or Windows installer, run from source with Node 20.6 or later, or use Docker Compose. Providers using openai-completions or anthropic-messages endpoints are configured in a JSON file. Permission modes and an optional OS-level sandbox gate tool actions. The architecture is deliberately single-user: one process and one active agent session, with no multi-user concurrency or tenant isolation.

## Features
- Three-layer memory (L1 learner profile, L2 native wiki, L3 session recall)
- Proactive cron scheduler with natural-language job creation
- Feishu and WeChat personal IM channels
- Practice Lab with in-browser terminal and run records
- Simple Mode with one-click preset workspaces
- Skill system and content hub for importing skills and presets
- Conversation branch editing
- Pluggable openai-completions and anthropic-messages providers
- Permission modes and optional OS-level sandbox
- Computer use in the desktop app
- Session replay showcase export
- Chinese/English UI with multiple themes

## Integrations
Pi coding-agent SDK, Feishu, WeChat, QQ, Anthropic, OpenAI, DeepSeek, Ollama, pi-sandbox, pi-computer-use, VeryMath, Docker

## Platforms
WINDOWS, MACOS, LINUX, WEB, API, CLI

## Pricing
Open Source

## Version
v0.7.0

## Links
- Website: https://hhyqhh.github.io/inno-agent-website
- Documentation: https://hhyqhh.github.io/inno-agent-website/getting-started/introduction/
- Repository: https://github.com/hhyqhh/inno-agent
- EveryDev.ai: https://www.everydev.ai/tools/inno-agent