Skip to content

Keeping an eye on things

Once agents are running on issues, you spend less time creating work and more time supervising it. This page covers the surfaces built for that: the Dashboard, Costs, the Inbox, and Activity — plus the live updates that tie them together.

Dashboard

Sidebar → Overview → Dashboard (also the default landing page when you open a company). Top to bottom:

Live agent status

A real-time strip of your agents and what they're doing right now — in-flight runs stream their latest output here. It updates over Server-Sent Events, so no refresh is needed.

Summary cards

Six at-a-glance metrics:

  • Total Agents — with an active / paused breakdown.
  • Open Issues — with how many were resolved this week.
  • Stale Issues — open longer than 14 days.
  • Heartbeat Success — the success rate of agent runs over the last 30 days.
  • Budget Spent (30d) — LLM spend this period, against your allocation if you've set one.
  • Approvals Pending — how many are waiting on a human.

LLM Cost Budget tile

If a monthly budget is set for the company, a tile shows month-to-date spend against it, with a colour band (green / warning / critical) and the percentage used. If no budget is set and you're an instance admin, you get a small form here to set one — see Costs below.

Heatmap, velocity, and goals

Below the cards: an Agent Activity Heatmap (last 14 days), an Issue Velocity chart (last 12 weeks), and Goal Progress bars for any goals you've defined. Good for spotting "things went quiet three days ago" patterns at a glance.

Recent Activity

A live feed of the latest events in the company, mirroring the full Activity log (below).

Costs

Sidebar → Company → Costs.

Spend breakdown

A cost table you can group By Agent or By Model, over a 7d, 30d, or 90d window. This is where you answer "which agent is burning the budget?".

Budget policies

The Costs page lists the company's budget policies — their scope, the metric and window they watch, and the threshold. Each policy is a guardrail: when spend crosses its threshold, Trimus opens a budget incident (below) rather than silently continuing.

Budget policies are created through the JSON API (POST /api/companies/{id}/budget-policies), not from a button on this page — they're typically set up by an operator or an agent. See the costs operator page and API.md.

Budget incidents

A log of every time a policy threshold was breached, with its status and when it opened and resolved. An admin can resolve an incident to acknowledge it.

The monthly LLM budget

Separately from policies, an instance admin can set a simple monthly LLM cost budget for a company from the Dashboard tile: a Monthly USD amount plus Warning % and Critical % thresholds. When month-to-date spend crosses a threshold, instance admins get a direct message. (This budget alerts; it does not by itself stop agents — see Pausing the world.)

Inbox

Sidebar → Overview → Inbox. Your "what's waiting on me" view, with three tabs:

  • Issues — actionable issues (assigned to you or needing attention).
  • Approvals — everything pending a human decision.
  • Failed Runs — recent agent runs that errored, so you can investigate.

Each item can be dismissed to hide it for now; dismissed items are restorable from the inbox's dismissed view. Items the system considers handled (an approval you resolved, say) drop off on their own.

Activity

Sidebar → Company → Activity. The full, append-only audit log for the company: every status change, comment, assignment, file write, and approval resolution, newest first. Filter by entity type or agent.

Reach for this when you need to reconstruct "what happened at 2pm Tuesday that broke things" — it's the canonical record. The same events show up scoped to a single issue on that issue's Timeline.

Live updates everywhere

Most pages subscribe to Server-Sent Events. New runs, comments, and status changes appear in place without a refresh. If data looks stale, your live connection may have dropped (a network blip or a proxy timeout) — refresh the page to reconnect.

Pausing the world

If you need to slow or stop agents:

  • Pause individual agents — click Pause on each agent. They stop picking up work but keep their assigned issues. Resume brings them back.
  • Turn off heartbeats — uncheck Enable automatic heartbeats on an agent's Configuration tab so it only runs when you trigger it by hand.
  • Stop the server — an operator action. It doesn't lose state; agents resume when the server comes back.

There is no single "pause every agent" button, by design — pausing agents one at a time is friction that prevents an accidental "I disabled the whole company" outage. Budget alerts notify you of overspend, but treat pausing agents as the deliberate way to halt work.