Give your AI assistant tools you control

Kamari is an open-source server for developers and self-hosters. It gives AI clients 370 permission-checked tools for your email, files, network and home.

MIT License | Version 1.35.15 | Source on Bitbucket

The Kamari dashboard's Pipelines page: a pipeline's owner, manual trigger and five numbered steps.

Your assistant acts; you stay in charge

It does real work

Send an email, read a PDF, check disk usage or provision a Proxmox VM. Each is one tool call, such as email_send or terraform_proxmox_provision.

You decide who may do what

Every call is checked against the caller's own permissions. An API key can never do more than the person who created it.

You can see what happened

Every tool call is written to an audit log. You can trace one job from its trigger to its last step.

Three steps from install to automation

  1. Install it and sign in

    You run one Python process. It prints a one-time admin token. You sign in with it, create your user and register a passkey.

    The Kamari sign-in screen with a passkey button and an API key field.
  2. Connect your AI client

    Kamari speaks MCP (Model Context Protocol), the standard AI clients use to call tools. Your client signs in through OAuth 2.1, and you pick the tools it may use.

    The consent page an MCP client opens: access presets, a tool filter, tools grouped by plugin, and Allow and Deny buttons.
  3. Automate and trace

    Pipelines run steps when an event arrives, such as a new email. Every run is recorded, and you can step through one in a debugger.

    The pipeline editor: a name, an email.received trigger and a condition step checking the subject contains Invoice.

Facts you can check

Counted from the code when this page was built.

What Kamari is not

  • Not a hosted service. You run it on your own machine. There is no account to create and no cloud tier.
  • Not an AI model. It gives tools to the AI client you already use. An optional plugin runs a local model through Ollama.
  • Not finished. 55 of 55 plugins are still marked Dev, and the code changes often. Read the changelog before you upgrade.

Install it in four steps

You need Python 3.12 or newer and git. Kamari is developed on Windows and runs in production on Linux.

  1. Get the code

    git clone https://bitbucket.org/Spac3k/kamari.git
    cd kamari
  2. Install it with every plugin's dependencies

    python -m venv .venv
    .venv/bin/pip install -e ".[all]"

    On Windows, use .venv\Scripts\pip instead of .venv/bin/pip.

  3. Start it

    .venv/bin/python -m kamari --data-dir ./data --host localhost

    On Windows, use .venv\Scripts\python. When it worked, the terminal shows a box like this:

    ============================================================
      Kamari admin token (valid until this process stops):
    
      (a random token)
    
      Authorization: ApiKey (the same token)
    ============================================================
  4. Sign in

    Open http://localhost:8000/web, choose Sign in with API Key and paste the token. Then create your user on the Internal page and register a passkey from its invite link.