OpenOcta — An Open-Source Personal Desktop AI Agent

Website · 简体中文

_{Double-click install · run locally · instant boot · your data stays on your machine}

OpenOcta is an open-source personal desktop AI agent. Double-click to install on your PC — an Agent that runs on your own machine under your full control. Use natural language for office work, IT ops, marketing, business analysis, and software testing.

Latest v1.0.5 (2026-07-02) — all-platform installers · fix for tool-call chat history · single Go binary with embedded Control UI

All releases →

🖥️ Personal desktop — built for your PC: files, terminal, browser, and local projects — not just a chat window.

⚡️ Double-click ready — install in ~30 seconds with a full Control UI out of the box — no hours of CLI setup.

🪶 Ultra-lightweight — ~30MB installers and low runtime memory — ideal for 24/7 background use.

🔒 Local-first — sessions and memory stay on your machine; data stays under your control and can run on intranet — no Node / Python in production.

🔧 Fully self-developed Go — Gateway, Agent, and Channels are native Go — single binary with embedded Control UI.

🔌 Skills & MCP — built-in tools + MCP protocol + skills marketplace — extend digital employee capabilities on demand.

💬 IM remote control — drive tasks from WeChat, WeCom, DingTalk, and Feishu anytime, anywhere.

🧠 Four-tier memory + L4 evolution — Knowledge Vault learns your preferences over time (docs).

Open source from China — maintained by a Chinese team, 100% Apache-2.0 — China's first open-source personal desktop AI agent.

OpenOcta is personal desktop-first: double-click install, full client UI, domestic IM, and CN LLM integrations. Versus OpenClaw (Node self-host) and Hermes (Python CLI), OpenOcta ships a fully self-developed Go runtime as a single binary, with L4 evolution and Knowledge Vault.

Full comparison & selection guide → docs/compare-openclaw-hermes.en.md

A quick look at the OpenOcta desktop client:

Agent chat · multi-turn collaboration, tool use, and complex tasks driven by natural language	Employee market · one-click install Zabbix, Prometheus, MySQL DBA, and more digital employees
Skills library · 766+ Skills across DevOps / databases / dev tools	Knowledge vault · Obsidian-compatible notes, synced index for Agent semantic search
Models · public and local model providers supported	Tutorials · video and doc courses for Windows / Mac / Linux quick start and ops practice

The client also includes Tools, Scheduled tasks, and IM channels. Download at openocta.com.

Office work — automate repetitive tasks and free up creativity.
e.g. “Summarize this week’s emails and draft a status report”

IT ops — 24/7 intelligent on-call, from firefighting to prevention.
e.g. “Analyze this alert log and suggest remediation”

Marketing — produce multi-channel content faster.
e.g. “Write a short social post from this product brief”

Knowledge management — local knowledge base with semantic search.
e.g. “Find XX policy in my vault and summarize it”

Dev collaboration — read code, edit repos, run commands, wire toolchains.
e.g. “Read this repo’s README and list items to fix”

More at openocta.com/cases and docs/scenarios.md.

1. Download — GitHub Releases or openocta.com
2. Install & open — desktop client, ready in ~30 seconds
3. Chat — describe a task in Chat; the agent uses files, terminal, and tools

_{Product demo · chat → skills → knowledge → ops agent (pending)}

Connect a model before your first chat:

Option A — UI (recommended)
Open the Models tab or follow the setup wizard on first launch, then enter your API key and choose a model.

Option B — Config file

Minimal example (Moonshot CN; see model providers for DeepSeek, Qwen, etc.):

{
  "env": {
    "vars": {
      "MOONSHOT_API_KEY": "sk-your-key"
    }
  },
  "agents": {
    "defaults": {
      "model": { "primary": "moonshot-cn/kimi-k2.5" }
    }
  }
}

Restart the client, then try in Chat:

List files in the current working directory and tell me the 3 largest

OpenOcta takes design cues from OpenClaw Gateway protocol and Control UI, and is fully reimplemented in Go.

Visit openocta.com/download:

Config is initialized on first run:

On macOS, install from the .dmg into Applications — see deploy/dist-README.md.

Links

• Website: https://openocta.com
• Quick start: platform guides on the download page
• Skills / MCP marketplace: https://resource.openocta.com
• Enterprise (AMC): https://amc.openocta.com

For self-hosting Gateway and Control UI (end users can use the installers above without building).

• Go 1.24+
• Node ≥18 (frontend build only — not required in production)
• ANTHROPIC_API_KEY (for agent CLI)

make build
./openocta gateway run

Gateway listens on http://127.0.0.1:18900 by default. The Control UI is embedded via go:embed — open that URL in a browser.

./openocta gateway run    # terminal 1
make run-ui               # terminal 2 → http://localhost:5173

export ANTHROPIC_API_KEY=your-key
./openocta agent -m "Hello, echo test"

• README.md

Upstream: OpenClaw · docs.openclaw.ai

OpenOcta/
├── src/          # Go backend
├── ui/           # Control UI
├── deploy/       # Installers, Docker, systemd
├── docs/         # Documentation
└── imgs/         # Logo, screenshots, community QR
    ├── readmePIC/    # README feature screenshots
    └── screenshots/  # Demo GIF and UI screenshots

• Star the repo: ⭐ Star and Watch for updates
• Good first issues: Issues for newcomers
• Contribute: CONTRIBUTING.md — Pull Requests welcome
• Issues: GitHub Issues
• Community chat: scan the QR code below or visit openocta.com

_{Scan to join the OpenOcta community group}

This project is licensed under ** Apache-2.0**.