openakita 1.7.0

全能自进化AI Agent - 基于Ralph Wiggum模式，永不放弃

OpenAkita

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

Setup Center • Features • Quick Start • Architecture • Documentation

中文文档

---

What is OpenAkita?

OpenAkita is a self-evolving AI Agent framework. It autonomously learns new skills, performs daily self-checks and repairs, accumulates experience from task execution, and never gives up when facing difficulties — persisting until the task is done.

Like the Akita dog it's named after: loyal, reliable, never quits.

• Self-Evolving — Auto-generates skills, installs dependencies, learns from mistakes
• Never Gives Up — Ralph Wiggum Mode: persistent execution loop until task completion
• Growing Memory — Remembers your preferences and habits, auto-consolidates daily
• Standards-Based — MCP and Agent Skills standard compliance for broad ecosystem compatibility
• Multi-Platform — Setup Center GUI, CLI, Telegram, Feishu, DingTalk, WeCom, QQ

---

Setup Center

OpenAkita provides a cross-platform Setup Center desktop app (built with Tauri + React) for intuitive installation and configuration:

• Python Environment — Auto-detect system Python or install embedded Python
• One-Click Install — Create venv + pip install OpenAkita (PyPI / GitHub Release / local source)
• Version Control — Choose specific versions; defaults to Setup Center version for compatibility
• LLM Endpoint Manager — Multi-provider, multi-endpoint, failover; fetch model lists + search selector
• Prompt Compiler Config — Dedicated fast model endpoints for instruction preprocessing
• IM Channel Setup — Telegram, Feishu, WeCom, DingTalk, QQ — all in one place
• Agent & Skills Config — Behavior parameters, skill toggles, MCP tool management
• System Tray — Background residency + auto-start on boot, one-click start/stop
• Status Monitor — Live service status dashboard with real-time log viewing

Download: GitHub Releases

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

---

Features

Feature	Description
Self-Learning & Evolution	Daily self-check (04:00), memory consolidation (03:00), task retrospection, auto skill generation, auto dependency install
Ralph Wiggum Mode	Never-give-up execution loop: Plan → Act → Verify → repeat until done; checkpoint recovery
Prompt Compiler	Two-stage prompt architecture: fast model preprocesses instructions, compiles identity files, detects compound tasks
MCP Integration	Model Context Protocol standard, stdio transport, auto server discovery, built-in web search
Skill System	Agent Skills standard (SKILL.md), 8 discovery directories, GitHub install, LLM auto-generation
Plan Mode	Auto-detect multi-step tasks, create execution plans, real-time progress tracking, persisted as Markdown
Multi-LLM Endpoints	9 providers, capability-based routing, priority failover, thinking mode, multimodal (text/image/video/voice)
Multi-Platform IM	CLI / Telegram / Feishu / DingTalk / WeCom (full support); QQ (implemented)
Desktop Automation	Windows UIAutomation + vision fallback, 9 tools: screenshot, click, type, hotkeys, window management
Multi-Agent	Master-Worker architecture, ZMQ message bus, smart routing, dynamic scaling, fault recovery
Scheduled Tasks	Cron / interval / one-time triggers, reminder + task types, persistent storage
Identity & Memory	Four-file identity (SOUL / AGENT / USER / MEMORY), vector search, daily auto-consolidation
Persona System	8 role presets (default / business / tech / butler / girlfriend / boyfriend / family / Jarvis), layered persona architecture (preset + user preferences + context-adaptive), LLM-driven trait mining
Living Presence	Proactive engine: greetings, task follow-ups, memory recall; frequency control, quiet hours, feedback loop; feels like a real assistant
Sticker Engine	ChineseBQB integration (5700+ stickers), keyword search, mood mapping, per-persona sticker strategy
Tool System	11 categories, 50+ tools, 3-level progressive disclosure (catalog → detail → execute) to reduce token usage
Setup Center	Tauri cross-platform desktop app, guided wizard, tray residency, status monitoring

---

Self-Learning & Self-Evolution

The core differentiator: OpenAkita doesn't just execute — it learns and grows autonomously.

Mechanism	Trigger	Behavior
Daily Self-Check	Every day at 04:00	Analyze ERROR logs → LLM diagnosis → auto-fix tool errors → generate report
Memory Consolidation	Every day at 03:00	Consolidate conversations → semantic dedup → extract insights → refresh MEMORY.md
Task Retrospection	After long tasks (>60s)	Analyze efficiency → extract lessons → store in long-term memory
Skill Auto-Generation	Missing capability detected	LLM generates SKILL.md + script → auto-test → register and load
Auto Dependency Install	pip/npm package missing	Search GitHub → install dependency → fallback to skill generation
Real-Time Memory	Every conversation turn	Extract preferences/rules/facts → vector storage → auto-update MEMORY.md
Persona Trait Mining	Every conversation turn	LLM analyzes user messages → extract personality preferences → daily promotion to identity
User Profile Learning	During conversations	Identify preferences and habits → update USER.md → personalized experience

---

Quick Start

Option 1: Setup Center (Recommended)

The easiest way — graphical guided setup, no command-line experience needed:

1. Download the installer from GitHub Releases
2. Install and launch Setup Center
3. Follow the wizard: Python → Install OpenAkita → Configure LLM → Configure IM → Finish & Start

Option 2: PyPI Install

# Install
pip install openakita

# Install with all optional features
pip install openakita[all]

# Run setup wizard
openakita init

Optional extras: feishu, whisper, browser, windows

Option 3: Source Install

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

Run

# Interactive CLI
openakita

# Execute a single task
openakita run "Create a Python calculator with tests"

# Service mode (IM channels)
openakita serve

# Background daemon
openakita daemon start

# Check status
openakita status

Recommended Models

Model	Provider	Notes
claude-sonnet-4-5-*	Anthropic	Default, balanced
claude-opus-4-5-*	Anthropic	Most capable
qwen3-max	Alibaba	Strong Chinese support
deepseek-v3	DeepSeek	Cost-effective
kimi-k2.5	Moonshot	Long-context
minimax-m2.1	MiniMax	Good for dialogue

For complex tasks, enable Thinking mode by using a *-thinking model variant (e.g., claude-opus-4-5-20251101-thinking).

Basic Configuration

# .env (minimum configuration)

# LLM API (required — configure at least one)
ANTHROPIC_API_KEY=your-api-key

# Telegram (optional)
TELEGRAM_ENABLED=true
TELEGRAM_BOT_TOKEN=your-bot-token

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                          OpenAkita                               │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  ┌──────────────────── Setup Center ────────────────────────┐   │
│  │  Tauri + React Desktop App · Install · Config · Monitor   │   │
│  └──────────────────────────────────────────────────────────┘   │
│                              │                                   │
│  ┌──────────────────── Identity Layer ──────────────────────┐   │
│  │  SOUL.md · AGENT.md · USER.md · MEMORY.md                │   │
│  │  Personas (8 presets + user_custom)                       │   │
│  └──────────────────────────────────────────────────────────┘   │
│                              │                                   │
│  ┌──────────────────── Core Layer ──────────────────────────┐   │
│  │  Brain (LLM) · Identity · Memory · Ralph Loop             │   │
│  │  Prompt Compiler · Task Monitor                           │   │
│  │  PersonaManager · TraitMiner · ProactiveEngine            │   │
│  └──────────────────────────────────────────────────────────┘   │
│                              │                                   │
│  ┌──────────────────── Tool Layer ──────────────────────────┐   │
│  │  Shell · File · Web · MCP · Skills · Scheduler            │   │
│  │  Browser · Desktop · Plan · Profile · IM Channel          │   │
│  │  Persona · Sticker                                        │   │
│  └──────────────────────────────────────────────────────────┘   │
│                              │                                   │
│  ┌──────────────────── Evolution Engine ────────────────────┐   │
│  │  SelfCheck · Generator · Installer · LogAnalyzer          │   │
│  │  DailyConsolidator · TaskRetrospection                    │   │
│  └──────────────────────────────────────────────────────────┘   │
│                              │                                   │
│  ┌──────────────────── Channel Layer ───────────────────────┐   │
│  │  CLI · Telegram · Feishu · WeCom · DingTalk · QQ          │   │
│  └──────────────────────────────────────────────────────────┘   │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Core Components

Component	Description
Brain	Unified LLM client, multi-endpoint failover, capability routing
Identity	Four-file identity system, compiled to token-efficient summaries
Memory	Vector memory (ChromaDB), semantic search, daily auto-consolidation
Ralph Loop	Never-give-up execution loop, StopHook interception, checkpoint recovery
Prompt Compiler	Two-stage prompt architecture, fast model preprocessing
Task Monitor	Execution monitoring, timeout model switching, task retrospection
Evolution Engine	Self-check, skill generation, dependency install, log analysis
Skills	Agent Skills standard, dynamic loading, GitHub install, auto-generation
MCP	Model Context Protocol, server discovery, tool proxying
Scheduler	Task scheduling, cron / interval / one-time triggers
Persona	3-layer persona architecture, 8 presets, LLM-driven trait mining, runtime state persistence
Proactive Engine	Living presence mode: proactive greetings, task follow-ups, memory recall, feedback-driven frequency control
Sticker Engine	ChineseBQB sticker integration, keyword/mood search, per-persona sticker strategy
Channels	Unified message format, multi-platform IM adapters

---

Documentation

Document	Description
Quick Start	Installation and basic usage
Architecture	System design and components
Configuration	All configuration options
Deployment	Production deployment (systemd / Docker / nohup)
MCP Integration	Connecting external services
IM Channels	Telegram / Feishu / DingTalk setup
Skill System	Creating and using skills
Testing	Testing framework and coverage

---

Community

Join our community for help, discussions, and updates:

WeChat Group Scan to join (Chinese)

WeChat — Chinese community chat Discord — Join Discord X (Twitter) — @openakita Email — [e-mail]

• Documentation — Complete guides
• Issues — Bug reports & feature requests
• Discussions — Q&A and ideas
• Star us — Show your support

---

Acknowledgments

• Anthropic Claude — LLM Engine
• Tauri — Cross-platform desktop framework for Setup Center
• browser-use — AI browser automation
• AGENTS.md Standard — Agent behavior specification
• Agent Skills — Skill standardization specification
• ZeroMQ — Multi-agent inter-process communication
• ChineseBQB — Chinese sticker/emoji pack collection

License

MIT License — See LICENSE

This project includes third-party skills licensed under Apache 2.0 and other open-source licenses. See THIRD_PARTY_NOTICES.md for details.

---

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