OpenAkita Deployment Guide (English)

OpenAkita Deployment Guide (English)

中文版

Complete deployment guide covering PyPI installation, source installation, LLM configuration, and IM channel setup

Table of Contents

• System Requirements
• Installation Methods
  • Method 1: PyPI Install (Recommended)
  • Method 2: One-Click Deploy Script
  • Method 3: Source Install
• Configuration
  • Configuration Files Overview
  • Environment Variables (.env)
  • LLM Endpoint Configuration (llm_endpoints.json)
  • IM Channel Configuration
  • Identity Configuration (identity/)
  • Memory System Configuration
  • Multi-Agent Orchestration
• Starting Services
• Publishing to PyPI
• Production Deployment
• FAQ
• Upgrading & Uninstalling

System Requirements

Hardware

Software

Supported Operating Systems

• ✅ Windows 10/11
• ✅ Ubuntu 20.04/22.04/24.04
• ✅ Debian 11/12
• ✅ CentOS 8/9 Stream
• ✅ macOS 12+

Installation Methods

Method 1: PyPI Install (Recommended)

The simplest way to get started:

# 1. Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/macOS
# or .\venv\Scripts\activate  # Windows

# 2. Install OpenAkita (core)
pip install openakita

# 3. Install optional features
pip install openakita[feishu]     # + Feishu (Lark) support
pip install openakita[whisper]    # + Voice recognition
pip install openakita[browser]    # + Browser AI agent
pip install openakita[windows]    # + Windows desktop automation
pip install openakita[all]        # Install all optional features (Windows-only deps are auto-skipped on non-Windows)

# 4. Run setup wizard
openakita init

# 5. Start
openakita

Method 2: One-Click Deploy Script

There are two one-click paths:

• One-click install (PyPI): fastest way to get a working installation (recommended)
• One-click deploy (Source): for development / modifying the repo

Method 2-A: One-click install (PyPI, recommended)

Linux/macOS:

curl -fsSL https://raw.githubusercontent.com/openakita/openakita/main/scripts/quickstart.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/openakita/openakita/main/scripts/quickstart.ps1 | iex

For extras / mirrors, download and run with parameters (recommended):

curl -fsSL -o quickstart.sh https://raw.githubusercontent.com/openakita/openakita/main/scripts/quickstart.sh
bash quickstart.sh --extras all --index-url https://pypi.tuna.tsinghua.edu.cn/simple

irm https://raw.githubusercontent.com/openakita/openakita/main/scripts/quickstart.ps1 -OutFile quickstart.ps1
.\quickstart.ps1 -Extras all -IndexUrl https://pypi.tuna.tsinghua.edu.cn/simple

The script installs into ~/.openakita/app and uses an isolated venv at ~/.openakita/venv by default (Windows: %USERPROFILE%\.openakita\...), to avoid polluting system Python.

Method 2-B: One-click deploy (Source)

Automatically installs Python, Git, dependencies, and everything else (requires cloning the repo first):

Linux/macOS:

git clone https://github.com/openakita/openakita.git
cd openakita
chmod +x scripts/deploy.sh
./scripts/deploy.sh

Windows (PowerShell):

git clone https://github.com/openakita/openakita.git
cd openakita
.\scripts\deploy.ps1

The script will automatically:

1. Detect and install Python 3.11+
2. Detect and install Git
3. Create virtual environment
4. Install dependencies (auto-fallback to Chinese mirror if needed)
5. Optionally install Playwright browsers
6. Optionally download Whisper voice model
7. Initialize .env and data/llm_endpoints.json
8. Create all required data directories
9. Verify installation
10. Optionally create systemd service (Linux)

Method 3: Source Install

# 1. Clone repository
git clone https://github.com/openakita/openakita.git
cd openakita

# 2. Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/macOS
# or .\venv\Scripts\activate  # Windows

# 3. Upgrade pip
pip install --upgrade pip

# 4. Install project (development mode)
pip install -e ".[all,dev]"

# 5. Install Playwright browsers (optional)
playwright install chromium

# 6. Copy configuration files
cp examples/.env.example .env
cp data/llm_endpoints.json.example data/llm_endpoints.json

# 7. Edit configuration
# Edit .env to fill in API Keys and IM channel settings
# Edit data/llm_endpoints.json to configure LLM endpoints

# 8. Run setup wizard (or configure manually)
openakita init

# 9. Start
openakita

Configuration

Configuration Files Overview

project-root/
├── .env                          # Environment variables (API Keys, IM Tokens, etc.)
├── data/
│   └── llm_endpoints.json        # LLM multi-endpoint config (models, priority, capability routing)
└── identity/
    ├── SOUL.md                   # Agent core personality
    ├── AGENT.md                  # Agent behavior specification
    ├── USER.md                   # User profile (auto-learned)
    └── MEMORY.md                 # Core memory (auto-updated)

Configuration priority: Environment variables > .env file > Code defaults

Environment Variables (.env)

Copy the example file and edit:

cp examples/.env.example .env

Required

# At least one LLM API Key is required
ANTHROPIC_API_KEY=sk-your-api-key-here

Tip: If you don't use Anthropic, you can configure other API Keys (e.g. DASHSCOPE_API_KEY) as long as they are properly referenced in data/llm_endpoints.json.

Full Environment Variable Reference

LLM Endpoint Configuration (llm_endpoints.json)

This is the core configuration file of OpenAkita, supporting multi-endpoint, automatic failover, and capability-based routing.

Configuration Methods

Method A: Interactive Wizard (Recommended)

python -m openakita.llm.setup.cli

The wizard supports:

• Selecting from known provider list
• Automatically fetching available models
• Testing endpoint connectivity
• Setting priorities
• Saving configuration

Method B: Manual Edit

cp data/llm_endpoints.json.example data/llm_endpoints.json
# Then edit the file

Configuration Structure

{
  "endpoints": [
    {
      "name": "claude-primary",          // Endpoint name (unique identifier)
      "provider": "anthropic",           // Provider identifier
      "api_type": "anthropic",           // API protocol: anthropic or openai
      "base_url": "https://api.anthropic.com",  // API base URL
      "api_key_env": "ANTHROPIC_API_KEY",       // API Key env variable name
      "model": "claude-opus-4-5-20251101-thinking",
      "priority": 1,                     // Priority (1 = highest)
      "max_tokens": 8192,               // Max output tokens
      "timeout": 60,                     // Timeout (seconds)
      "capabilities": ["text", "vision", "tools"],  // Capability declaration
      "extra_params": {},                // Extra API parameters
      "note": "Anthropic Official API"   // Note
    }
  ],
  "settings": {
    "retry_count": 2,                    // Retries per endpoint
    "retry_delay_seconds": 2,            // Retry delay (seconds)
    "health_check_interval": 60,         // Health check interval (seconds)
    "fallback_on_error": true            // Auto-failover to backup endpoint
  }
}

Field Reference

Capability Routing

When a user sends an image, the system automatically selects an endpoint with vision capability; for video, it selects one with video capability.

Failover Mechanism

1. Endpoints are tried in priority order (lowest first)
2. On failure, automatically switches to the next endpoint
3. Failed endpoints enter a 3-minute cooldown period
4. Endpoints automatically recover after cooldown

Provider Configuration Examples

Anthropic (Claude)

{
  "name": "claude",
  "provider": "anthropic",
  "api_type": "anthropic",
  "base_url": "https://api.anthropic.com",
  "api_key_env": "ANTHROPIC_API_KEY",
  "model": "claude-sonnet-4-20250514",
  "priority": 1,
  "capabilities": ["text", "vision", "tools"]
}

Qwen (DashScope / Alibaba Cloud)

{
  "name": "qwen",
  "provider": "dashscope",
  "api_type": "openai",
  "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
  "api_key_env": "DASHSCOPE_API_KEY",
  "model": "qwen3-max",
  "priority": 2,
  "capabilities": ["text", "tools", "thinking"],
  "extra_params": {"enable_thinking": true}
}

Kimi (Moonshot AI)

{
  "name": "kimi",
  "provider": "moonshot",
  "api_type": "openai",
  "base_url": "https://api.moonshot.cn/v1",
  "api_key_env": "KIMI_API_KEY",
  "model": "kimi-k2.5",
  "priority": 3,
  "capabilities": ["text", "vision", "video", "tools"],
  "extra_params": {"thinking": {"type": "enabled"}}
}

DeepSeek

{
  "name": "deepseek",
  "provider": "deepseek",
  "api_type": "openai",
  "base_url": "https://api.deepseek.com/v1",
  "api_key_env": "DEEPSEEK_API_KEY",
  "model": "deepseek-chat",
  "priority": 4,
  "capabilities": ["text", "tools"]
}

OpenRouter (Multi-model Aggregator)

{
  "name": "openrouter-gemini",
  "provider": "openrouter",
  "api_type": "openai",
  "base_url": "https://openrouter.ai/api/v1",
  "api_key_env": "OPENROUTER_API_KEY",
  "model": "google/gemini-2.5-pro",
  "priority": 5,
  "capabilities": ["text", "vision", "video", "tools"]
}

MiniMax (Anthropic Protocol)

{
  "name": "minimax",
  "provider": "minimax",
  "api_type": "anthropic",
  "base_url": "https://api.minimaxi.com/anthropic",
  "api_key_env": "MINIMAX_API_KEY",
  "model": "MiniMax-M2.1",
  "priority": 6,
  "capabilities": ["text", "tools"]
}

Using a Proxy / Relay Service

If direct access to Anthropic is difficult, use a relay service:

{
  "name": "claude-proxy",
  "provider": "anthropic",
  "api_type": "anthropic",
  "base_url": "https://your-proxy-domain.com",
  "api_key_env": "ANTHROPIC_API_KEY",
  "model": "claude-sonnet-4-20250514",
  "priority": 1,
  "capabilities": ["text", "vision", "tools"]
}

IM Channel Configuration

OpenAkita supports 5 major IM platforms, all enabled via .env:

Telegram

1. Create a Bot at @BotFather and get the Token
2. Configure .env:

TELEGRAM_ENABLED=true
TELEGRAM_BOT_TOKEN=123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11
# Users in mainland China must configure a proxy
TELEGRAM_PROXY=http://127.0.0.1:7890

3. On first use, the Agent generates a pairing code in data/telegram/pairing/ (visible in console output)

Feishu (Lark)

1. Create an app at Feishu Open Platform
2. Enable Bot capability and add message-related permissions
3. Configure .env:

FEISHU_ENABLED=true
FEISHU_APP_ID=cli_xxxxx
FEISHU_APP_SECRET=xxxxx

4. The Feishu adapter uses WebSocket long connection by default (recommended) — no callback URL needed

WeCom (WeWork)

1. Create an internal app at WeCom Admin Console
2. Get Corp ID, Agent ID, and Secret
3. Configure .env:

WEWORK_ENABLED=true
WEWORK_CORP_ID=ww_xxxxx
WEWORK_AGENT_ID=1000002
WEWORK_SECRET=xxxxx

DingTalk

1. Create an internal app at DingTalk Open Platform
2. Enable Bot capability
3. Configure .env:

DINGTALK_ENABLED=true
DINGTALK_CLIENT_ID=dingxxxxx
DINGTALK_CLIENT_SECRET=xxxxx

QQ Official Bot

Create a bot at QQ Open Platform and get credentials:

QQBOT_ENABLED=true
QQBOT_APP_ID=your-app-id
QQBOT_APP_SECRET=your-app-secret
QQBOT_SANDBOX=false

OneBot (Universal Protocol)

Requires a running OneBot implementation (e.g. NapCat):

ONEBOT_ENABLED=true>

# Mode 1: CLI + IM (interactive mode with IM channels running simultaneously)
openakita

# Mode 2: IM-only service (background service, no CLI)
openakita serve

# Create from example files
cp identity/SOUL.md.example identity/SOUL.md
cp identity/AGENT.md.example identity/AGENT.md
cp identity/USER.md.example identity/USER.md
cp identity/MEMORY.md.example identity/MEMORY.md

# Configure in .env
EMBEDDING_MODEL=shibing624/text2vec-base-chinese  # Recommended for Chinese
EMBEDDING_DEVICE=cpu                                # Set to cuda if GPU available

python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('shibing624/text2vec-base-chinese')"

pip install torch --index-url https://download.pytorch.org/whl/cu118
# Set EMBEDDING_DEVICE=cuda in .env

# Configure in .env
ORCHESTRATION_ENABLED=true
ORCHESTRATION_BUS_ADDRESS=tcp://127.0.0.1:5555
ORCHESTRATION_PUB_ADDRESS=tcp://127.0.0.1:5556
ORCHESTRATION_MIN_WORKERS=1
ORCHESTRATION_MAX_WORKERS=5

openakita           # Interactive CLI (with IM channels running simultaneously)
python -m openakita # Same

openakita serve     # IM-only service, no CLI interaction

openakita run "Analyze the code structure of the current directory"

openakita init              # Run setup wizard
openakita status            # Show Agent status
openakita selfcheck         # Run self-check
openakita compile           # Compile identity files (reduces token usage)
openakita prompt-debug      # Show prompt debug info
openakita --version         # Show version

# 1. Install build tools
pip install build twine

# 2. Build package
python -m build

# 3. Check package
twine check dist/*

# 4. Upload to PyPI
twine upload dist/*
# Or upload to TestPyPI
twine upload --repository testpypi dist/*

# 1. Update version in pyproject.toml
# 2. Create tag
git tag v1.2.2
git push origin v1.2.2
# 3. GitHub Actions automatically builds and publishes to PyPI

# Install from PyPI
pip install openakita

# Verify
openakita --version
python -c "import openakita; print(openakita.__version__)"

[Unit]
Description=OpenAkita AI Agent Service
After=network.target

[Service]
Type=simple
User=your-user
WorkingDirectory=/path/to/openakita
Environment="PATH=/path/to/openakita/venv/bin"
ExecStart=/path/to/openakita/venv/bin/openakita serve
Restart=always
RestartSec=10
StandardOutput=journal
StandardError=journal

[Install]
WantedBy=multi-user.target

sudo systemctl daemon-reload
sudo systemctl enable openakita
sudo systemctl start openakita
sudo systemctl status openakita

# View logs
journalctl -u openakita -f

FROM python:3.11-slim

WORKDIR /app

# Install system dependencies
RUN apt-get update && apt-get install -y git && rm -rf /var/lib/apt/lists/*

# Install Python dependencies
COPY pyproject.toml .
RUN pip install --no-cache-dir -e ".[feishu]"

# Copy project files
COPY . .

# Install Playwright
RUN playwright install chromium && playwright install-deps chromium

CMD ["openakita", "serve"]

docker build -t openakita .
docker run -d \
  --name openakita \
  -v $(pwd)/.env:/app/.env \
  -v $(pwd)/data:/app/data \
  -v $(pwd)/identity:/app/identity \
  openakita

source venv/bin/activate
nohup openakita serve > logs/serve.log 2>&1 &
echo $! > openakita.pid

python --version
# Windows: py -3.11 -m venv venv
# Linux: pyenv install 3.11.8 && pyenv local 3.11.8

# Use Chinese mirror (for users in China)
pip install openakita -i https://pypi.tuna.tsinghua.edu.cn/simple
# Or set permanent mirror
pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple

# Linux: install system dependencies
playwright install-deps
# Or install only Chromium
playwright install chromium

# Use CPU-only PyTorch (saves ~2GB)
pip install torch --index-url https://download.pytorch.org/whl/cpu
# Use a smaller embedding model
# EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2

# Use the interactive tool to test
python -m openakita.llm.setup.cli
# Choose "4. Test endpoint" to verify connectivity

# PyPI install
pip install --upgrade openakita

# Source install
cd openakita
git pull
pip install -e ".[all]"

# Stop service
sudo systemctl stop openakita
sudo systemctl disable openakita
sudo rm /etc/systemd/system/openakita.service

# Uninstall package
pip uninstall openakita

# Remove data (use caution)
rm -rf data/ identity/ logs/