Agent settings
The gear in the submenu header opens a page with nine tabs. Heavy content loads only for the active one.
Insights
Section titled “Insights”The agent’s durable conclusions: solutions found, new links, feature ideas, candidates for skills and automations, verification results — each with evidence and a confidence level, plus Confirm and Dismiss.
This is not an event feed: one-off occurrences don’t land here.
Agent runs
Section titled “Agent runs”The history of background work: ordinary runs and research. Badges on top show how many are queued, running, completed, failed and cancelled.
A run shows its type, status, step progress, start and finish time, the original task, the acceptance criteria, the plan and — on failure — a “why the run failed” block. The final answer is delivered to the originating chat, with an “Open chat” button, and “Open the resulting note” when the result was a note.
Automations
Section titled “Automations”Scheduled agent actions: statistics (total, active, paused) and the Active and Paused lists.
A rule shows its schedule in plain words (“At 9:00 daily”), the last run time and any error text. Buttons: Run now, Enable / Disable, Edit, Delete.
The rule form: a name, a cron schedule with a hint (“0 9 * * *” — every day at 9:00), an analyser (calendar, birthdays, mail, finance, tasks, wiki, a cross-domain digest, or “Custom scenario (skill)”), the skill for a custom scenario, and an optional refinement.
The other way is conversational: “New automation” runs a three-question interview, after which the agent creates the rule itself.
The web agent
Section titled “The web agent”What the agent is doing in the browser and how to step in. A built-in engine — no external programs.
- Take control — pause the agent and enter a login, captcha or 2FA code yourself;
- Give back to the agent;
- Pause / Resume;
- Close.
Requests over 7 days, input and output tokens, cache hit rate, error count, and a table of recent calls: time, task, provider, model, tokens, latency.
Action history
Section titled “Action history”A technical log: which action the agent attempted, whether confirmation was required, how long it took and how it ended. Columns: time, action, status, risk, decision, duration.
This is where to go with “who changed this” when the change wasn’t made by a human.
The agent factory
Section titled “The agent factory”The system spots repeating work and builds specialised agents. Sections: Specialists, Discovery, Evaluation, Lifecycle.
The operating mode is switched explicitly:
| Mode | What it does |
|---|---|
| Legacy: read-only | the registry is readable, changes are stopped |
| v2: manual | you create and check specialists by hand |
| v2: discovery | background collection, no drafts created |
| v2: full loop | at most two drafts per cycle, promotion through gates |
Switching modes changes behaviour only: collected data, evaluations and drafts are preserved. A specialist can also be built by hand through a six-step wizard.
Skills and the wiki
Section titled “Skills and the wiki”Skills and the knowledge base have their own page.