Staff the job automatically
The Assistant breaks down the request, checks the team, and recruits the specialists the work needs.
One-man company OS
Run a one-man company with AI employees — from a single chat window.
imboss is a desktop app where you chat with an AI Assistant — your chief of staff. Tell it what you need, and it assembles a team of AI employees, hands out the work in group chats, and reports back with the result.
View on GitHub
macOS · Windows · Linux Local-first — your data stays in
~/.imboss Bring your own model provider Free for personal use · Enterprise licensing available Open source on GitHub Why imboss
One person, run like a company
You don't need to hire to scale. Give imboss one clear request, and it organizes the plan, the people and the work — then shows you exactly how the result was made.
See how it worksEmployees you actually configure
Every employee has its own role, model, thinking level, skills, MCP servers and schedule.
Work you can audit
Each task runs in a group chat with ask cards and progress cards, so nothing is a black box.
How it works
You talk. The Assistant runs the company.
Five steps from a plain request to a finished result.
- 01
You talk to the Assistant
Chat naturally, like messaging a colleague — no prompt engineering required.
- 02
The Assistant plans and staffs the job
It breaks your request into tasks, checks who is already on the team, recruits specialists when the right expertise is missing, and opens a task group chat.
- 03
Employees do the work
Each employee has its own role, persona, model, tools, skills and MCP servers. They research, write, run commands and produce real deliverables.
- 04
The team coordinates itself
Employees @-mention each other, ask questions and report results, while the Assistant keeps the task moving and unblocks them.
- 05
You stay in control
Follow progress in the chat, answer questions when a decision is genuinely needed, and receive the finished result.
Key features
Everything you need to run the work
A chat-first workspace, a team that ships, and full control over models, tools and data.
Chat-first workspace
A WeChat-style messaging UI with conversation list, avatars, attachments, unread badges and read state.
Rich messages
Markdown with syntax highlighting and KaTeX math, plus ask cards for structured answers and live progress cards.
An Assistant that manages the team
The built-in Assistant plots the plan, recruits employees, creates task chats, assigns work and reports to you. Edit its persona and system prompt anytime.
A team of employees
Recruit specialists with a name, role and persona. Give each one its own model, thinking level, skills, MCP servers and optional scheduled tasks.
Task group chats
Every task gets a group chat. See the full back-and-forth between your employees, so you can audit how any result was produced.
Model providers
Connect Anthropic, OpenAI, DeepSeek, Google, Groq, MiniMax, Moonshot AI, OpenRouter, xAI, Xiaomi, Z.ai and GitHub Copilot — or add a custom provider.
MCP servers
Register Model Context Protocol servers over stdio or streamable HTTP, and assign them per employee so each session only connects the tools it needs.
Skills
Install Agent Skills from local folders (SKILL.md is the source of truth) and choose exactly which skills each employee can use.
Mini Apps
Run local web apps in the sidebar next to your chats — a static page or a process you start. They keep running as you move between modules.
Scheduled tasks
Give an employee a cron schedule to wake it up automatically — a daily digest or a periodic check. See the next run time and toggle tasks per employee.
Messaging integrations
Bridge imboss to WeChat (official iLink Bot), QQ (official bot platform) and Telegram (@BotFather) so you can work from your phone.
Notifications
Native desktop notifications when an agent posts while the window is in the background — click to jump straight to the chat.
Appearance and language
Light, dark and system themes with 18 accent colors, in English and Simplified Chinese.
Private and local by design
Everything lives in ~/.imboss — database, avatars, agent environment, sessions and mini apps. API keys stay local with restrictive file permissions.
Extend it your way
Bring your own tools, skills and apps
Connect the tools you already trust
Register Model Context Protocol servers over stdio or streamable HTTP, and assign them per employee — each session only connects the tools it actually needs, with per-server and app-wide environment variables.
Give employees the right skills
Install Agent Skills from local folders, where SKILL.md is the source of truth, and choose exactly which skills each employee can use. The Assistant ships with built-in skills to manage MCP servers, manage skills and build mini apps for you.
Run mini apps right beside the chat
A mini app is just a folder with a metadata.json — a static page, or a process you start with its own command, args, working directory and environment. It keeps running as you move between modules, and the Assistant can build and debug it for you.
Works with your stack
Your models. Your messaging apps.
Connect a curated list of model providers — or bring a custom one. Then bridge imboss to the chat apps on your phone.
Model providers
Connect Anthropic, OpenAI, DeepSeek, Google, Groq, MiniMax, Moonshot AI, OpenRouter, xAI, Xiaomi, Z.ai and GitHub Copilot — or add a custom provider. Supports OpenAI-compatible completions/responses, Anthropic Messages and Google Generative AI APIs. Fetch available models with one click; API keys stay masked until you reveal them.
- Anthropic
- OpenAI
- DeepSeek
- Groq
- MiniMax
- Moonshot AI
- OpenRouter
- xAI
- Xiaomi
- Z.ai
- GitHub Copilot
- Custom
Work from your phone
Bridge imboss to WeChat (via the official iLink Bot), QQ (via the official bot platform) or Telegram (via a @BotFather bot). Owner accounts are whitelisted: messages you send are treated as your instructions, and the Assistant's replies come right back to that conversation. Inbound images and files are imported into the chat.
- Telegram
See it in action
A chat window that runs a team
The Assistant assigns work, employees report back, and progress cards keep you posted — all inside one conversation.
Get started
Download imboss
Available for macOS, Windows and Linux. Pick your platform and grab the latest release from GitHub.
Free for personal use · Enterprise licensing: kefroth@outlook.com
View all releases on GitHub
After launch, add a model provider, then say hello to your Assistant in Chats.
Private by design
Your work stays on your machine
- Everything imboss creates lives in a single folder in your home directory: ~/.imboss — database, avatars, agent environment, sessions and mini apps.
- imboss manages its own isolated Pi agent environment, so it never touches a system-wide Pi installation.
- API keys are stored locally with restrictive file permissions.
FAQ
Questions, answered
What is imboss?
A desktop app where you chat with an AI Assistant who assembles a team of AI employees, assigns the work in group chats and reports back with results.
Do I need my own API key?
Yes. imboss connects to the model providers you configure (Anthropic, OpenAI, DeepSeek, Google and more) or a custom provider. Add one on first launch.
Where is my data stored?
In a single local folder, ~/.imboss, on your machine. API keys stay local with restrictive file permissions.
Which platforms are supported?
macOS, Windows and Linux. Download the installer for your platform and it is ready to run.
Do I need Node.js to use it?
No — only if you build from source. Building requires Node.js ≥ 22, pnpm ≥ 9 and build tools for native modules (Xcode CLT on macOS, MSVC on Windows, build-essential on Linux).
Can I use it from my phone?
Yes — bridge imboss to WeChat, QQ or Telegram. Messages from the owner account are treated as your instructions, and replies are delivered back to that conversation.
Is imboss free?
imboss is free for personal use. Enterprise use requires a paid license — email kefroth@outlook.com for licensing details.
Is imboss open source?
Yes — the source code is on GitHub, released under the GPL-3.0 license.