open-super-agent
Self-hosted, always-on AI teammate with its own computer, an open-source Grok Bot alternative that works with your choice of model.
At a Glance
MIT-licensed self-hosted core product; users bring their own model and pay model providers per token.
Engagement
Available On
Alternatives
Listed Nov 2026
About open-super-agent
open-super-agent is an open-source, self-hosted AI teammate published by kadaliao under the MIT license. Each bot chats, delegates long tasks to background jobs, and uses a computer that the user can watch and take over. It runs as a macOS desktop app, a Docker Compose server, an iPhone app in TestFlight beta, or inside chat apps. The README labels the project status as alpha, and the latest listed release is v0.17.0.
What It Is
The project is a self-hosted agent platform in the style of Grok Bot. Each bot has its own persona, memory and conversations. Time-consuming work goes to background tasks while the chat continues, and several bots can be placed in a group so that one researches and hands results to another for summarizing.
The Bot's Computer
Bots browse, fill in forms and download files on a computer that the user can view live. When a step needs a CAPTCHA or login, the bot asks the user to take over. Parallel tasks each get their own browser window. The computer can be the bot's own (local Docker sandbox or the Instabox cloud computer), or the user's Mac: with permission, the bot controls apps via mouse and keyboard and stops when the user moves the mouse.
Models, Apps and Skills
Models are interchangeable: DeepSeek, Kimi, Zhipu, OpenAI, local Ollama, or any OpenAI-compatible endpoint, with cost tracking per task and day and a daily budget. An app marketplace lists 40 apps such as Notion, Gmail, GitHub, Feishu, DingTalk and Tencent Docs, and arbitrary MCP servers can be connected. Users can demonstrate a task once so the bot turns it into a skill invoked with a slash command; skills can include scripts and references, be installed from a skill marketplace, or imported from Claude SKILL.md files. Scheduled tasks can be created in plain language.
Safety and Deployment
Actions such as sending email or submitting forms prompt for approval first, and secrets are stored encrypted in a vault. All data can be exported as one encrypted file. By default there is no login and it is meant for local or private-network use; for public deployment the README says to set OSA_PASSWORD and use HTTPS. Bots can be reached from Telegram, Feishu, DingTalk, WeCom and Discord, and the interface is available in Chinese and English.
Community Discussions
Be the first to start a conversation about open-super-agent
Share your experience with open-super-agent, ask questions, or help others learn from your insights.
Pricing
Open Source (Self-hosted)
MIT-licensed self-hosted core product; users bring their own model and pay model providers per token.
- MIT License
- Self-hosted via Docker Compose or macOS desktop app
- Bring your own model (DeepSeek, Kimi, OpenAI, Ollama, any OpenAI-compatible API)
- Model usage billed per token by the chosen provider (external cost)
- Status: alpha
Capabilities
Key Features
- Persistent bots with personas, memory and conversations
- Background tasks and multi-bot group collaboration
- Live-viewable bot computer with user takeover
- Local Mac control via mouse and keyboard with permission
- App marketplace with 40 apps and arbitrary MCP support
- Approval prompts before sensitive actions
- Encrypted secret vault
- Switchable models including any OpenAI-compatible endpoint
- Per-task and per-day cost tracking with daily budget
- Teach-by-demonstration skills with Claude SKILL.md import
- Scheduled tasks from natural language
- Chat channel access via Telegram, Feishu, DingTalk, WeCom, Discord
- Encrypted full data export
- Chinese and English interface
