What runs where, what leaves, and who decides.
Short on purpose. Every line below is how the platform works today, not a roadmap — the roadmap is at the bottom, labelled.
Where it runs.
Each company gets its own isolated instance and its own storage — one container, one volume, nothing shared between customers. On our infrastructure in the EU by default; in your own cloud on request. Nothing is installed on your machines; the instance connects out to Slack, your board and your repository with the tokens you grant.
What leaves your tools.
To the AI model
To us
Meeting audio
Credentials.
Tokens pasted in chat are caught deterministically before any model sees them, stored encrypted in your instance, and never repeated. Each integration gets the minimum scope — we list the exact permissions to grant. He never sees a secret value: secrets are masked in everything he reads, including test output.
Who decides what he may do.
Supervised
Delegated (default)
Autonomous
Set per kind of change by an owner in the portal. Merging, production and connecting new tools require a human yes at every level; production access is read-only unless separately documented and approved, and a production-changing action always requires human confirmation. Areas you mark restricted — an admin panel, a billing page — he cannot open: the rule is enforced in code before any request leaves. Every change is listed in the thread with a link and logged, and designed to be reversible where the connected system supports reversal.
The record.
Every action he takes is logged in your instance. The project memory — decisions, promises, what was tested where — is a git repository you can read. The weekly report is counted from that log, not written from memory. An owner can see, at any moment, which account and model he runs on, how much of the plan is used, and what he is doing right now.
Subprocessors.
Anthropic (the model, under your account) · Slack, and Microsoft where Teams is used · Hetzner Online GmbH, Germany (hosting of the instances) · OpenAI, only when voice in calls is enabled · Vercel and Google for the website and the portal sign-in. Your own connected tools act under your authorisation and are not our subprocessors. The full list with purposes is in the privacy policy; we notify you before adding one.
Honest status.
SOC 2 Type 1
Data processing agreement
Independent penetration test
We'd rather say so than imply otherwise.
Questions your security team will ask. We'll answer them on a call.
Bring the checklist. We bring the architecture, the data flows and the logs.