How Monitor works
Monitor is the Portal's main screen. It shows what your agents did, what their work cost, and how Ekamyth Agent OS's checks are doing, from the records Agent OS keeps on your machine.
How does Monitor work?
Monitor is one screen in your browser that reads the records Agent OS keeps on your machine: model calls through the model gateway, tool-call decisions from the hooks, sessions and checks. You open it with hero portal open, pick a tab, and click into any figure to see what's behind it. It shows only your own activity.
Monitor concepts
- Portal. The Agent OS screens in your browser, served from your machine. Monitor is one of its three views, with Control for your own work and the Brain page. See
hero portal. - Tab. One question Monitor answers, such as Health or Telemetry. See The tabs.
- Panel. One figure, chart or table on a tab.
- Drill-in. Clicking a figure to see the records behind it. See Step by step.
- You. Monitor shows only your own activity. See What Monitor shows.
How you use Monitor, step by step
- Open it. Run
hero portal openon the machine where Agent OS runs. Your browser opens the Portal, signed in. - Start on Overview. Monitor opens on the Overview tab unless the address names another tab.
- Pick a tab for the question you have. See The tabs.
- Drill into a figure to see the records it was built from. The detail view has its own tabs: Usage, Routing, Reliability, Outcomes, Evidence, Governance and Traces.
- Choose a time window. Overview offers
24h,7dand30d. Telemetry offers Live snapshot, This week, Last week and Last month.
The tabs on Monitor
| Tab | What it answers | What it shows |
|---|---|---|
| Overview | Maturity, value and attention | Your use against your allowance, and what was saved and could still be saved. |
| Health | Is each capability working right now | One tile per capability. |
| Telemetry | What your agents' jobs did | Search everything, and a list of jobs. Each row: a readable title, who and which session, how it ended, and how it was routed. Below it are the detail views: Usage, Routing, Reliability, Outcomes, Evidence, Governance and Traces. |
What Monitor shows, and what it leaves out
- Only you. Monitor shows only your own activity.
- Only what's recorded. A panel whose data can't be read says so and names the read that failed.
- Some panels are left out. Outcomes and Evidence show a note instead of figures, and the Usage detail view leaves out use by stage.
Monitor data, and what leaves your machine
Monitor reads the records Agent OS already keeps on your machine; see Data files. It runs at 127.0.0.1, so it's reachable only from this machine. Monitor's own tabs only read those records. Other Portal features can send data: a chat in the Portal passes your messages and system context through the model gateway to the model you picked, and testing a provider key contacts that provider. For what Agent OS as a whole sends, see How Privacy works.
When part of Monitor fails
| What fails | What you see | What to do |
|---|---|---|
| The Portal isn't running | hero portal open: no running portal to open — start it: hero portal | Run hero portal, or hero portal --install-service. |
| The Portal's files are missing | hero portal exits 3 and says to reinstall. | Reinstall Agent OS. |
What you can change in Monitor
| Setting | Default | Where |
|---|---|---|
| The Portal's port | 7878 | hero portal --port or HERO_PORTAL_PORT |
| Keep the Portal running after a restart | On, from the installer | hero portal --install-service |
| The time window | 7d on Overview, and This week on Telemetry | On the screen |
See Monitor for yourself
hero portal --service-status
hero portal open
Monitor limits and next steps
| Limit | Behaviour | Can you change it? | What to do |
|---|---|---|---|
| Outcomes and Evidence | A note, not figures. | No | Use hero trace for what each session did. |
| Whose activity | Only your own. | No | None. |
| Where you can open it | Only on this machine. | No | None. |
| Counts that differ from the commands | Rule coverage can show 0 while hero enforce lists your rules. The pending-approvals count can differ from hero inbox. Governance decision tiles and block counts can differ from the CLI. The Enforce table can show 0 for a rule that has already stopped calls. Monitor can say nothing is enforcing your budget while Control says Enforced. | No | Trust the command: hero inbox for approvals, hero attest rules for block counts, hero govern status for enforcement, and Control for your budget. |
✓ governed in Recent receipts | The tick means Agent OS allowed the call, not that it succeeded. A call the provider rejected, for example at a rate limit, can still show it. | No | Trace and Telemetry show the call as failed. |
| Open trace | Monitor's Open trace can show no steps while hero trace shows them. | No | Use hero trace show <session>. |
Checked against Agent OS 0.1.36rc14 (build 599c5e120c18) on 17 Nov 2026.