Meta Tech ProviderSee What Your AI Assistant Did: An Activity Log for Every API Key
The short version
Under Settings → API keys → What AI assistants did, ThinnestAI lists the changes assistants made in your workspace through the MCP server — and every time one was stopped to ask for your approval — newest first, with the key used, the tool and how it went. The page shows the latest 25; the last 200 per key are kept.
Why it matters
Handing an assistant a key is handing it a set of hands. Approvals stop the dangerous moves; the activity log answers the question that comes after: what did it actually do? When a teammate asks why an agent's instructions changed on Tuesday, or who resolved a conversation, the answer is one page away.
What each line shows
| Shown | Example |
|---|---|
| When | 5 Oct, 14:02 |
| Which key | The key's name, so you know which assistant or teammate it was |
| What it did | update agent, create action, pause campaign |
| How it went | Done, Failed, or Asked for approval |
Only tools that change something are logged. Reads are not, and neither are test chats — dozens of them would bury the real changes.
What it deliberately does not record
What the assistant was asked to do, and the details it passed — a customer's phone number, a message's text — are not stored. The log is an audit of actions, not a copy of your customers' data. If you need the details of a change, the change itself is where it was made: the agent's settings, the conversation, the campaign.
The key is still the boundary
The log tells you what happened; the key decides what can. A Build key cannot message a customer or place a call at all, and a Read-only key cannot change anything. To end an assistant's access, revoke its key in Settings → API keys — it stops immediately. Give each assistant its own key and the log tells you exactly which one did what. More in giving an assistant a key that cannot message customers.