WorkingRoom

WorkingRoom

People and AI, working safely in the same place.

WorkingRoom is an open-source shared workspace for teams and AI agents. It brings files, documents and internal knowledge together in one place, and AI reads, writes, searches and organizes them within the permissions people set.

View on GitHub　｜　Documentation

The WorkingRoom screen. A chat where an AI agent proposes an operation on a shared file and a person approves it

AI agents work on shared files only after a person approves. Apache License 2.0 / Self-hosted / Latest release v0.2.7 (September 15, 2026)

Why WorkingRoom

AI models are improving rapidly. But in real work, a smart model is not enough. What really matters are these four points:

  • What information the AI can access
  • Whose permissions an operation runs with
  • What was changed
  • How to prevent mistakes and unintended operations

WorkingRoom treats AI not as a simple chatbot but as "a member of the team". The AI reads and edits files, organizes information and carries out tasks. At the same time, it works under the same access control as human members, and people always have the final say.

Key features

🗂️ Shared workspace

Members of the same tenant and AI agents share the same file system. Each member also has a private folder that only they and the AI acting with their permissions can use.

🔐 Permission boundaries

Access control is allowlist-based, so nothing is accessible by default. You grant read / write permissions on folders and files per access group.

✅ Human approval flow

Before the AI creates, edits, deletes or moves a file, an approval card with the diff is shown. No change is applied until you press "Approve".

🤖 Custom AI agents

Create agents with a system prompt, a model tier (light / medium / heavy) and a working folder, and use them again and again.

🔌 MCP and Slack integration

Connect external MCP servers (Streamable HTTP) to add tools to your agents. Connect Slack with OAuth2 to read and send messages.

🏠 Open source and self-hosted

Released under the Apache License 2.0. Each organization runs its own instance, so data is separated at the infrastructure level.

People and AI share the same data space

In WorkingRoom, users and AI agents access the workspace through the same chat screen. The AI works through file operation tools, and every operation is subject to access permissions and the approval flow.

WorkingRoom concept diagram. People and AI working together through a shared workspace

The diagram below shows the architecture. Both users and AI agents use the chat UI, and the web app reads and writes files and folders in storage through the file operation tools.

Mermaid
flowchart TB
  subgraph Actors[Users]
    User[User]
    Agent[AI agent]
  end

  subgraph WebServer[Web server]
    WebApp[WorkingRoom web app]

    subgraph Workspace[Shared workspace]
      Chat[Chat UI]
      FileTools[File operation tools]
      LocalStorage[(Local storage)]
    end
  end

  User <--> Chat
  Agent <--> Chat

  Chat --> WebApp
  WebApp --> FileTools
  FileTools <--> LocalStorage

  LocalStorage --- Files[Files / folders]

Supported AI providers

ProviderEnvironment variables
Anthropic (Claude)ANTHROPIC_API_KEY
OpenAIOPENAI_API_KEY
Google (Gemini)GOOGLE_GENERATIVE_AI_API_KEY
Self-hosted LLM (OpenAI-compatible API)SELF_HOSTED_BASE_URL, SELF_HOSTED_API_KEY

You need an API key for at least one AI provider. Tenant owners can fix the provider used across the tenant on the "Settings" page.

AI agents can work with images (jpeg, png, webp, gif), PDF and text (text-based files such as text, md and csv).

Getting started

With Docker, you can start WorkingRoom right away.

  1. Set one or more of ANTHROPIC_API_KEY, OPENAI_API_KEY and GOOGLE_GENERATIVE_AI_API_KEY as shell environment variables.
  2. Get the repository and run the following commands. Database migrations run automatically at startup.
  3. Open http://localhost:3000 in your browser.
Bash
git clone https://github.com/kairo-inc/working-room.git
cd working-room
docker compose up

For local development, you need Node.js 20–24 and Yarn 4 or later. See the README in the repository for the steps.

WorkingRoom is in early development (v0.2.x). The API and the database schema may change before v1.0.

Roadmap

The main features in development or planned. They may change depending on priorities.

TimingFeatures
In developmentRestoring deleted files (trash)
PlannedPasting files from the clipboard, folder upload, file search (full text and file name), more file formats such as spreadsheets and presentations, AWS S3 storage
FutureMulti-tenant operation, sign-in with a Google account

FAQ

Is WorkingRoom free to use?

Yes. WorkingRoom is open-source software under the Apache License 2.0, and you can self-host it in your own environment for free. However, API usage fees for AI providers (Anthropic, OpenAI, Google and others) are charged separately according to each provider's pricing.

Can the AI rewrite files on its own?

Operations such as creating, editing, deleting and moving files are shown as approval cards in the chat before they run, and are not applied until the user presses "Approve". In chats with "Auto-approve" turned on, approval can be skipped, but even then the AI cannot do anything beyond the user's own permissions.

What should I do with files I don't want other members to see?

Put them in your private folder. Files in a private folder can be accessed only by that user and the AI agents acting with that user's permissions.

Can I use an in-house LLM or an on-premises model?

Yes. You can use a self-hosted LLM that provides an OpenAI-compatible API by setting the environment variables SELF_HOSTED_BASE_URL and SELF_HOSTED_API_KEY.

How do I update when a new version is released?

If you run WorkingRoom with Docker, update it with the following steps. Database migrations run automatically at startup, so you don't need to run them manually.

  1. Stop WorkingRoom with docker compose stop.
  2. Back up your data. In the Docker setup, the database (SQLite) and the workspace files are stored in the workspace-data folder at the root of the repository. Copy the whole folder.
  3. Fetch the latest release tag and switch to it. The default branch of the repository (dev) contains code in development, so specify a release tag.
  4. Rebuild the image and start it.
Bash
git fetch --tags
git checkout v0.2.7
docker compose up -d --build

Replace v0.2.7 with the version you are updating to. You can check the changes in each version in CHANGELOG.md. The database schema of WorkingRoom may change until v1.0, so always back up before updating.

How can I contribute to development?

See CONTRIBUTING.md in the GitHub repository.

The standard workspace for a future where AI is part of every team.

GitHub repository　｜　Documentation (workingroom.io)