A self-hosted AI email assistant for Outlook and Gmail

Outlook, Gmail, Teams, Slack and Google Chat stay in separate tabs, and Triazz triages them into one stream and a “while you were away” digest. Each message is sorted by project, urgency, whether it needs a reply, and risk (money, legal, HR, credentials, unknown sender). Rules sit under the model: the model can add risk flags but never remove them.

A reply is drafted in your voice under a Guard, and waits in a queue inside Triazz. Drafts are not saved to the Drafts folder in Gmail or Outlook. There are also reminders and follow-ups for things you are waiting on.

Triazz starts in Shadow Mode, which reads and drafts only. Risky abilities such as sending email or chat are off by default. Per action you choose Do, Ask or Never, and a kill switch stops everything outward. Going Live is your own confirmed choice, and by default every bot reply opens with a line saying it writes on your behalf.

Every step is written to a tamper-evident activity log. See the Security & Trust Centre for what it does and does not protect against.

Triazz is self-hosted: your mail, the database and your connection credentials stay on the computer or server you run it on. What leaves is the specific text sent to the AI model you chose for a given job and the calls to the services you connect. With a local Ollama model, no text goes to a model provider. See the Privacy page for the details.

You choose a provider and model for triage, drafting and tasks separately: Ollama, Anthropic or OpenAI, or stay on the local default.

• Mail and chat are polled, with optional instant push for Teams, Slack and Google Chat. Attachments are not read.
• We recommend starting with one pilot user and widening from there.
• Triazz needs Node 22+ (and Ollama for local models), and is set up with us because the source is not public.
• Teams, Slack and Google Chat, tasks, browser QA and repository answers are part of the product too, but this page is about email.