Privacy policy
This policy describes what Cabinet stores when it is installed in a Slack workspace, which model providers see your text, and how long data is kept. Cabinet is run by agntwrk. Last updated 2026-10-12.
What Cabinet stores
Messages addressed to Cabinet: mentions, direct messages, replies in threads it takes part in, slash commands, and the reactions and button presses you give its replies. For every model call it stores a transcript: the system prompt, the messages, tool results, and the answer, with the model, token counts, cost, and any fallbacks. Decisions, meeting minutes, ideas, and retros are written as markdown to a private ledger repository on GitHub. It also stores your Slack workspace and user ids, the bot token (encrypted), the name and email returned by Sign in with Slack, and standard server logs (IP address, user agent, request path).
What Cabinet reads
Only messages in channels Cabinet has been added to or joined and in direct messages with it. It does not read other channels, does not sell Slack data, and does not use it to train models.
Model providers
Prompts, which contain your messages and the context needed to answer, may be processed by Anthropic (Claude), OpenAI (ChatGPT and the API), and Google (Gemini on the Gemini API and Vertex AI), depending on which lane a persona is set to use. Web research may send search queries to the provider's search tool. Each provider handles the data under its own terms for API and developer products.
Where it lives
Google Cloud in the United States (Cloud Run, Firestore, Cloud Storage, Cloud Tasks), a private GitHub repository for the ledger, and Slack.
Analytics
None. Cabinet uses no advertising or analytics trackers.
Retention and deletion
Sessions, turns, and transcripts are kept while Cabinet is installed. Delivery and sign-in records expire within days, and inbound Slack items after 30 days. The ledger stays in its repository until the owner deletes it. Write to us and we delete a workspace's data, normally within 30 days.