Where do I find the runs of a Tares agent, including failed ones, and see how each one ended?

Tares keeps every run of an agent under the agent’s page, in the Runs list, with status, rounds, duration and errors. Failed and inconclusive runs are recorded there and never appear on an entity’s timeline. For an agent that concludes, an outcome column shows how each run ended: its verdict, no verdict, or no finding. You can filter the list to successful or failed runs and use Show more to go past the first fifty.

How it works

Each time a trigger wakes a Tares agent, Tares records a run. The run belongs to the project whose wiring started it, so an agent’s page inside a project shows that project’s runs.

The Runs list on the agent’s page gives you, for every run:

• Status, so you can tell a finished run from a failed one.
• Rounds, shown as rounds against the cap the run was held to (rounds/max).
• Duration.
• Errors, if the run failed.
• Outcome, for an agent that concludes. It shows the run’s verdict, no verdict, or no finding, with the headline beside it.

A run that is still open shows the same outcome above its finding.

Only a run that produces a conclusion writes a finding to the entity’s timeline. A run that concludes “inconclusive” or fails stays in the Runs list and never pollutes a timeline. This keeps the timeline limited to real conclusions, and the Runs list holds the full record of what the agent did.

A run can also end as exhausted. This happens when the model runs out of rounds mid-investigation and still cannot conclude after one final call with tools disabled. The run then keeps the last text as a partial note and tells you to raise max rounds. The cap is 1 to 24 rounds, set under Advanced on the agent form or as max_rounds in catalog YAML. The default is 6, or 12 once the agent uses external MCP servers.

Set it up

You do not configure anything to get run history. It is recorded for every agent. To see it in action:

1. Start Tares locally if it is not running. The bundled demo ships an agent called incident-first-look, so you have a run to look at.

tares up

2. Fire the agent’s trigger. In the demo, inject a fault. The agent must be on in the project, because new agents start off.
3. Open the project’s Overview. The conclusion appears under What the agents found, with its headline, summary and next step.
4. Open the agent’s page and go to Runs. Read status, rounds, duration and errors for each run, and the outcome column for how it ended.
5. Use the filter to show only successful or only failed runs.
6. Click Show more to page past the first fifty runs.
7. To debug a firing, open the trigger’s page under Firings, or Advanced setup’s Activity. The run the firing woke is nested under it.

The filter and Show more use this request, which you can also call yourself:

GET /api/agents/builtin/{name}/runs?status=ok&offset=50

The status value filters to successful or failed runs, and offset=50 starts after the first fifty. Replace the name placeholder with your agent’s name.

The Build an AI SRE guide walks the whole loop on a demo system, from fault to firing to run to finding.

What you get

• One place to check every run of an agent, successful or not, with status, rounds, duration and errors.
• A clear answer to how each run ended: its verdict, no verdict, or no finding, with the headline next to it.
• A clean timeline. Inconclusive and failed runs stay in the Runs list, so the entity’s timeline in Explore holds only findings that were written back.
• A filter for successful or failed runs, so you can go straight to what broke.
• Paging past the first fifty runs with Show more.
• Visibility into budget. Each run shows rounds/max, so you can see when an agent is hitting its cap and raise max rounds.
• A link from firing to run. The trigger’s Firings and the Activity view show the run each firing woke.

When something else fits

• To read what an agent concluded rather than how its runs ended, use the project’s Overview under What the agents found, or the entity’s timeline in Explore.
• To see why an agent did not run at all, look at the trigger’s Firings. Without a model provider, each firing is logged as “no model provider” and no run produces a finding.

Related