AI agent blocks and the audit log
Four blocks for products where agents do the work and people stay in charge, plus an integrations directory and an audit log.
New
- A new AI agents group opens the catalog beyond chat blocks, with four free blocks.
- Agent 1 shows a live run: the task and its status ("Waiting for you" is announced), the plan, every note, tool call and approval in order, Stop, Run again and a follow-up box.
- Agent 2 is an approvals inbox with risk, the full request beside the list, A and D to approve or deny, J and K to move, and one button for every low-risk request.
- Agent 3 lists runs with status, time, tokens and cost, and a waterfall of model, tool and retrieval calls in which a failed span opens first; Agent 4 compares one prompt on two models with
{{variables}}as fields, ⌘Enter to run, ratings and Pick this one. - Settings 4 is an integrations directory with Connect progress and failure reasons, and Disconnect after a confirmation.
- Settings 5 is an audit log of who changed what (people, agents, API keys, the system), filterable, with a before-and-after diff per event and a CSV export.
Improved
- All six have loading, empty and error states (the playground has running and error), pass every scorecard check and ship their tests.
- The MCP server finds them for agent, approval, trace, observability, playground, audit and integrations.