What Custodian includes
Four parts, all working today. Each one is described as it is, not as it might become.
Capture from the assistants your team uses
A browser extension for Chrome, Edge and Brave records chats on ChatGPT, Claude, Gemini and Microsoft Copilot. A small command-line connector does the same for coding tools: Claude Code, Codex CLI, GitHub Copilot CLI and Antigravity. Both start only when the employee switches them on, and can be paused at any time.
- Access to four assistant sites only, nothing else in the browser
- From coding tools, only what the person writes and what the assistant answers: no code, no files, no commands
- The text waits on the employee's machine; the employee can read it, remove a chat or exclude it for good
- Sent once a day by default, or when the employee chooses
A private dashboard for each participant
Each person sees their own trends in strain, clarity, engagement and confidence, and nobody else sees them.
- The states of this week and this month, and a prediction for the next turn
- How often the assistant's replies were rated safe
- Download everything, or erase the account and all its records, without asking anyone
A safety check on the assistant
Every assistant reply is rated safe, borderline or harmful. This includes the dismissive or hostile answers that general moderation filters are not built to catch.
- Finds 84% of the harmful replies in our held-out test
- An early warning, not a guarantee: no human reviews each rating
- Measured and explained on the How it works page
A group view for HR
HR sees how the people taking part are doing as a group, to decide on tools, workload and training. Never to look for a person.
- Only groups of five or more participants
- A state or a week is shown only if five or more people share it
- No names, accounts, conversations or individual scores
Limits we state up front
Custodian is an early-warning tool for organisations and a mirror for the people who take part. It is not a medical device, and it does not diagnose.
- No real-time alerts. Dashboards refresh at the interval of your plan: every month, week or day.
- No integration with HR systems, and no export of individual results to anyone but the participant.
- No human review of each rating. The safety check misses about one harmful reply in six.
- No conformity assessment under the EU AI Act yet. We go through the classification with you before deployment.
See it on your own conversations
Custodian starts with one team and grows from there. Set it up now, or write to us first.