carole.ai 0.1.4

www.caroleai.com

Carole.ai is a local AI agent workspace with chat, project tools, memory, MCP integrations, browser automation, and a Kanban board. The PyPI package includes the built web interface: one caroleai command runs the FastAPI backend and serves the frontend from the same address.

The public landing page is caroleai.com.

It does not host accounts or the backend: Sign in and Sign up open http://127.0.0.1:8000 on the visitor's own computer after they install and start Carole.ai.

Click the preview to watch the full video with sound.

Install and run

Use Python 3.11 or 3.12. A virtual environment is recommended:

python -m venv .venv
# macOS/Linux: source .venv/bin/activate
# Windows PowerShell: .venv\Scripts\Activate.ps1
python -m pip install carole.ai
caroleai

Open http://127.0.0.1:8000. The first account created becomes the local instance owner. Application data is stored in ~/.carole by default. Node.js and Docker are not required for the PyPI installation.

You can also install the command with pipx install carole.ai. To use browser automation, install the Chromium browser once with playwright install chromium. Connect an AI model provider in the app's settings or supply its API key through your environment.

To use Google Workspace tools, open Settings → Google Workspace Integration and choose Connect Google Account. Carole opens Google's consent page for the user's own account; no shared Google login is used. Refresh and access tokens remain on that computer in Windows Credential Manager, macOS Keychain, or Linux Secret Service. Read operations can run directly, while sending email or creating, editing, trashing, or deleting Google data requires human approval.

Run caroleai --help for options including --port, --data-dir, and --no-browser. The command binds to 127.0.0.1 by default. Keep it local unless you have deliberately secured access: the instance owner can use host-level capabilities such as shell, Git, plugins, and browser automation.

For persistent login sessions, set a stable, random JWT_SECRET in your environment or .env file before starting Carole.ai. Without one, a new secret is generated on each restart and existing sessions are invalidated. Never commit API keys or secrets to this repository.

Features

• Hybrid GraphRAG memory with LanceDB and a networkx code graph.
• MCP integrations and a dynamic tool registry.
• Gmail, Calendar, Meet, and Google Tasks integration through local OAuth.
• Browser and meeting workflows, including Google Meet integration.
• Real-time Kanban and project-management tools.

Develop from source

The public GitHub repository contains both the FastAPI backend and the Next.js frontend. The following is for contributors working on the source; it is not needed to run the PyPI package.

Prerequisites: Python 3.11 or 3.12 and Node.js 22. From the repository root, create a Python environment and install the backend requirements:

python -m venv .venv
# Activate .venv for your shell, then:
python -m pip install -r backend/requirements.txt
cd backend
uvicorn main:app --reload

In another terminal, start the frontend:

cd frontend
npm ci
npm run dev

The development frontend is at http://localhost:3000, the API at http://localhost:8000, and API documentation at http://localhost:8000/docs. Set provider API keys in your own local environment if you want to use model-backed features.

Source, bug reports, and contributions are available through GitHub and its issue tracker.

Instance owner and upgrade notes

The first account created is automatically the local instance owner. An operator can optionally set CAROLE_OWNER_ID to an account UUID as a recovery override. The instance owner controls host capabilities: terminal/shell access, Git, browser automation, plugins, MCP servers, external workspace paths, and instance settings/model/prompt administration. These operations are unavailable until an owner is configured. Project ownership alone does not grant access to the host machine. Host tools run with the backend OS account's privileges; this is not an OS sandbox for untrusted tenants.

File, search, Git, and history APIs require an owned project UUID. Clients must use the backend Git API; the former Next.js Git endpoint returns HTTP 410. Attachments require authenticated downloads. Google accounts must reconnect through Settings after upgrading: credentials are stored separately per account in the operating-system credential vault, and any legacy private token file is migrated and removed. The bundled public Desktop OAuth client ID identifies Carole.ai; no OAuth client secret is included in the repository or package. It grants no Google account access by itself—access begins only after the user consents locally through the PKCE-protected flow. The OAuth start request must include browser credentials so its callback can verify the initiating browser.

Docker stores application data in the carole_data volume at /data/carole. Back up and copy any existing .carole data into that volume before switching an existing deployment; creating the volume does not migrate old data. An explicit DATABASE_URL or CAROLE_HOME_DIR override must point to persistent storage. Playwright browsers are installed at a path available to the non-root app user.