Privacy Policy

Munim · last updated 4 September 2026

Munim is an open source command line tool that runs on your own computer. It is not a hosted service. There is no Munim server, no Munim account, and no operator of Munim who can see your data.

The short version. Your credentials stay on your machine, in ~/.munim/credentials.json, a file only your user account can read. Data from the services you connect goes between your computer and those services directly. Munim runs as an MCP server inside a coding agent, so what its tools return reaches whichever model that agent runs on. Munim's own reasoning features are switched off by default and send nothing to a model host until you turn them on. Nothing is ever sent to the authors of Munim.

What Munim handles

Credentials. When you connect a service, the resulting OAuth token or API key is stored in ~/.munim/credentials.json with file permissions 0600, meaning only your user account can read it, filed under an identifier for the client it belongs to. It is not encrypted: anything running under your user account can read it without asking, the same as ~/.ssh/id_rsa or ~/.aws/credentials. A backup or disk snapshot of your home directory therefore contains it in readable form, so treat those the way you treat the machine itself. Two kinds belong to the installation rather than to any client, and are filed under a reserved name instead: the OAuth application credential for Gmail and Stitch, and the API key for a model host if you configure one. Munim's code never returns a raw credential to the code that uses it; adapters receive an already authenticated connection.
Service data. Whatever the service returns while you are using it: DNS records, deployment details, project names. If you connect Gmail, that includes message content and metadata your instructions cause it to read, and drafts your instructions cause it to write.
Run logs and reports. A record of what was done, written to ~/.munim/runs and ~/.munim/reports on your computer.

Where it goes

To the service itself. Munim talks directly from your computer to each provider's own endpoint. Nothing is proxied.

To the coding agent you run Munim inside, always. Munim is an MCP server. Your coding agent calls its tools and places the results into its own model's context, so whatever a tool returns reaches whichever model provider that agent uses, under that agent's terms rather than Munim's. This is true of every tool and of every result: the list of clients, the DNS findings, anything a provider hands back. It is a property of how MCP works and Munim cannot switch it off. If that matters for a particular client's data, the decision is which coding agent you run, not which Munim settings you choose.

To a model host of your own, only if you turn it on. Munim can also build its own agent using the Strands Agents SDK, for the tools that explain a finding rather than measure one. This is off by default. Until you run munim config ai on, Munim constructs no model client and sends nothing to any model host, and the checks, the audit and the mail plan all still work, because they are deterministic and never needed one. When you do turn it on, it runs on a host you choose and configure: Amazon Bedrock, or Google Gemini or Anthropic with the matching optional install. The data that agent is working with is sent to that provider under your own API key and their terms. If Gmail is connected and you ask it to work with mail, mail content reaches that provider.

Nowhere else. Munim contains no analytics, no telemetry, no crash reporting and no update check. It sends nothing to its authors. You can confirm this by reading the source.

Google user data

Munim's use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.

Specifically: Google user data is used only to provide the features you invoke. It is not sold, not transferred to anyone except the coding agent you run Munim inside and, if you have turned Munim's own reasoning on, the model provider you configured for it, in order to perform those features. It is not used for advertising and not used to train any model by Munim. Munim's authors have no access to it.

Keeping and removing it

Everything lives on your computer for as long as you leave it there.

munim disconnect "<client>" <service> removes one stored credential.
munim disconnect --all removes every client's stored credentials. It lists them and asks first; add --yes in a script, or --dry-run to see the list without removing anything.
munim config ai unset <host> removes a stored model API key, and munim config app unset <provider> removes a stored OAuth application credential. These belong to the installation rather than to a client, so disconnect does not touch them.
Deleting ~/.munim removes everything: the credentials, the run logs, the reports and the client list. Every provider would need connecting again.
Revoking Munim's access at the provider, for example at Google account permissions, invalidates the credential regardless of what is on your machine.

Children

Munim is a tool for people administering their own or their clients' web infrastructure. It is not directed at children and collects nothing from anyone.

Changes and contact

Changes to this policy will be published on this page with a new date, and in the project's commit history, which is public.

Questions: open an issue at github.com/vishalsg42/munim/issues, or use the contact address on the consent screen.