#219 docs(privacy): disclose AI assistant connector session history

Adds an AI assistant connectors section to /privacy. It covers the AgentWorkforce/cloud#5082 change, which saves ChatGPT and Grok Bot connector activity to the workspace's Agent Relay session history.

The section says:

• What is stored: each Agent Relay tool call with its arguments and result, relay messages the assistant sends, and relay messages it reads. The rest of the assistant conversation is not stored.
• Why: so the workspace can see what the assistant did, search it, and pick up its work later.
• Safeguards: values that look like credentials (webhook URLs, keys, tokens, authorization headers) are removed first, and each record is size-capped.
• Who sees it: visible to members of the workspace's organization, like other uploaded sessions. OpenAI processes it for embeddings and analysis, as the existing service-provider paragraph already says.
• Retention: kept with the workspace's session history until deleted. No separate retention period is documented for History, so none is claimed.
• Stop and delete: stop recording by disconnecting the app or removing the connection. Delete by contacting the address on the page; there is no self-service delete for a single session yet.

Validation

• npm ci --ignore-scripts=false in a fresh worktree, then cd web && npx tsc --noEmit -p tsconfig.json: exit 0
• npx vitest run lib/test/privacy-page.test.tsx (new test): 1 passed. With the page change stashed it fails, so the test checks the new copy.

Needs the repo's two independent reviews and the human/cmo gates before merge and deploy. It should go live together with, or before, cloud#5082 reaching production.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

---

[!NOTE]
Low Risk
Public privacy-policy text and a static render test only; no runtime product or data-handling code changes.

Overview
Adds an AI assistant connectors section to the /privacy policy, ahead of “How we use information,” describing MCP-linked assistants (e.g. ChatGPT and Grok Bot).

The new copy explains what connector activity is written to workspace session history (tool calls with args/results, relay messages sent and read), that full assistant chat outside Agent Relay tools is not stored, credential-like values are stripped and records are size-limited, who can see it, OpenAI processing for embeddings/analysis, retention with other session history, and how to stop recording or request deletion.

A Vitest SSR test in web/lib/test/privacy-page.test.tsx asserts the page renders the key disclosure phrases so the copy cannot be removed accidentally.

^{Reviewed by Cursor Bugbot for commit 6c7a552dfe2431666f067c446d71f796b29e5fc3. Bugbot is set up for automated code reviews on this repo. Configure here.}

---

Summary by cubic

Adds an AI assistant connectors section to the privacy page disclosing that ChatGPT and Grok Bot connector activity is saved to the workspace's Agent Relay session history.

• Covers what is stored (tool calls with arguments and results, relay messages sent and read), credential redaction, size caps, visibility, retention, and how to stop or delete recording.
• Adds a test asserting the new disclosure text renders on the privacy page.

---

Agent Relay sessions

• claude session a74f8f49 · opened via gh pr create · last active 2026-10-14