Self-Evolving AI Agent — Learns Autonomously, Never Gives Up

Desktop Terminal • Features • Quick Start • Architecture • Documentation

English | 中文

An AI Agent that keeps getting smarter while you sleep.

Most AI assistants forget you the moment the chat ends. OpenAkita teaches itself new skills, fixes its own bugs, and remembers everything you've told it — like the Akita dog it's named after: loyal, reliable, never quits.

Set up in 3 minutes with just an API key. 8 personas, 6 IM platforms, and yes — it sends memes.

OpenAkita provides a cross-platform Desktop Terminal (built with Tauri + React) — an all-in-one AI assistant with chat, configuration, monitoring, and skill management:

• AI Chat Assistant — Streaming output, Markdown rendering, multimodal input, Thinking display, Plan mode
• Bilingual (CN/EN) — Auto-detects system language, one-click switch, fully internationalized
• Localization & i18n — First-class support for Chinese and international ecosystems, PyPI mirrors, IM channels
• LLM Endpoint Manager — Multi-provider, multi-endpoint, auto-failover, online model list fetching
• IM Channel Setup — Telegram, Feishu, WeCom, DingTalk, QQ Official Bot, OneBot — all in one place
• Persona & Living Presence — 8 role presets, proactive greetings, memory recall, learns your preferences
• Skill Marketplace — Browse, download, configure skills in one place
• Status Monitor — Compact dashboard: service/LLM/IM health at a glance
• System Tray — Background residency + auto-start on boot, one-click start/stop

Download: GitHub Releases

Available for Windows (.exe) / macOS (.dmg) / Linux (.deb / .AppImage)

No command line. No config files. From install to conversation in 3 minutes:

① Fill in  → Add LLM endpoint + IM (optional)
② One-click → Auto-create env, install deps, write config
③ Done      → Launch service, start chatting

Workspace → Python → Install → LLM Endpoints
→ IM Channels → Tools & Skills → Agent System → Finish

Switch between modes anytime — click "Switch Setup Mode" in the sidebar to return to the selection page without losing existing configuration.

See Configuration Guide for full details.

Other AIs forget you the moment you close the chat. OpenAkita self-evolves — while you sleep, it's learning:

Every day 03:00  →  Memory consolidation: semantic dedup, extract insights, refresh MEMORY.md
Every day 04:00  →  Self-check: analyze error logs → LLM diagnosis → auto-fix → report
After each task   →  Retrospection: analyze efficiency, extract lessons, store long-term
When stuck        →  Auto-generate skills + install dependencies — it won't be stuck next time
Every chat turn   →  Mine your preferences and habits — gets to know you over time

Example: You ask it to write Python, it finds a missing package — auto pip install. Needs a new tool — auto-generates a Skill. Next morning, it's already fixed yesterday's bugs.

For complex reasoning, enable Thinking mode — just add -thinking suffix to the model name (e.g., claude-opus-4-5-20251101-thinking).

The easiest way — download, drop in an API Key, click, done:

1. Download from GitHub Releases (Windows / macOS / Linux)
2. Install and launch OpenAkita Desktop
3. Choose Quick Setup → Add LLM endpoint → Click "Start Setup" → All automatic → Start chatting

Need full control? Choose Full Setup: Workspace → Python → Install → LLM → IM → Tools → Agent → Finish

pip install openakita[all]    # Install (with all optional features)
openakita init                # Run setup wizard
openakita                     # Launch interactive CLI

git clone https://github.com/openakita/openakita.git
cd openakita
python -m venv venv && source venv/bin/activate
pip install -e ".[all]"
openakita init

openakita                              # Interactive chat
openakita run "Build a calculator"     # Execute a single task
openakita serve                        # Service mode (IM channels)
openakita daemon start                 # Background daemon
openakita status                       # Check status

# .env (just two lines to get started)
ANTHROPIC_API_KEY=your-api-key     # Or DASHSCOPE_API_KEY, etc.
TELEGRAM_BOT_TOKEN=your-bot-token  # Optional — connect Telegram

Desktop App (Tauri + React)
    │
Identity ─── SOUL.md · AGENT.md · USER.md · MEMORY.md · 8 Persona Presets
    │
Core     ─── Brain(LLM) · Memory(Vector) · Ralph(Never-Give-Up Loop)
    │        Prompt Compiler · PersonaManager · ProactiveEngine
    │
Tools    ─── Shell · File · Web · Browser · Desktop · MCP · Skills
    │        Scheduler · Plan · Sticker · Persona
    │
Evolution ── SelfCheck · Generator · Installer · LogAnalyzer
    │        DailyConsolidator
    │
Channels ─── CLI · Telegram · Feishu · WeCom · DingTalk · QQ Official · OneBot

See Architecture Doc for full details.

WeChat Group Scan to join

WeChat — Chinese community Discord — Join Discord X (Twitter) — @openakita Email — zacon365@gmail.com

Issues · Discussions · Star

• Anthropic Claude — Core LLM engine
• Tauri — Cross-platform desktop framework
• ChineseBQB — 5700+ stickers that give AI a soul
• browser-use — AI browser automation
• AGENTS.md / Agent Skills — Open standards
• ZeroMQ — Multi-agent IPC

MIT License — See LICENSE

Third-party licenses: THIRD_PARTY_NOTICES.md

OpenAkita — Self-Evolving AI Agent That Sends Memes, Learns Autonomously, Never Gives Up