Live Operations
Operations is the live view of running work: host health, agent runs, log stream, token spend, and (when an agent builds a page) a sandboxed preview.
Open Operations in the sidebar.
What you seeβ
The page refreshes from a live stream about every two seconds:
- Host CPU, RAM, disk, and sensors.
- Active runs: status, step count, and who started them.
- Actions waiting for your approval.
- Token use for the day and month, by model.
Several browser tabs share the same snapshot. Opening extra tabs does not multiply load on the machine.
Log streamβ
The embedded terminal is read-only. It shows what agents are doing; it is not a shell.
| Tag | Color | Meaning |
|---|---|---|
[INFO] | Green | Normal progress. |
[TOOL] | Cyan | A tool call, its inputs, and how long it took. |
[WARN] | Yellow | Retry, rate limit, or fallback model. |
[ERROR] | Red | Failed step or sandbox block. |
To run commands yourself, use the Web Terminal.
Live previewβ
When an agent generates a web page or chart, Operations can show it in a sandboxed iframe. The preview only accepts relative paths or HTTPS/loopback URLs and does not send referrer headers.
Token spendβ
Use the cost panel to see:
- Daily and monthly token totals (and estimated USD).
- Share per provider (OpenAI, Anthropic, Google, local Ollama).
- Which agents or missions are using the most budget.
Related pagesβ
- Dashboard β approvals and agent state.
- Missions β the job those logs belong to.
- Notifications β errors and reconnect alerts.