Self-hosted AI assistant for teams
Give your team a proactive assistant — in Telegram, WhatsApp, and the browser — that runs entirely on your own servers. Per-user isolation, a tamper-evident audit trail, and approval gates on anything that touches the real world.
Request team accessCloud AI assistants ask you to ship your team's data to someone else's servers. Mitten keeps all of it on yours.
The controls a serious team needs before an assistant touches anything.
Mitten runs entirely inside your infrastructure. No third-party SaaS, no vendor processing your conversations — bring your own model and nothing ever leaves the perimeter.
Per-user data isolation enforced at the database with row-level security. One person's context never bleeds into another's.
Every action recorded in a hash-chained log. Prove what the assistant did, when, and for whom — for compliance and incident review.
Email + password with JWT sessions, an admin tier, and per-role tool access. Provision and de-provision people as the team changes.
Risky tool calls pause for explicit approval. Nothing destructive happens without a person signing off.
Per-user message and tool-call limits, with sensible defaults — so an automated agent can't run away with your model budget.
An annual self-hosted license for up to ten people. You host it, you control it, and your model spend goes straight to the provider.