What should I do when an AI SRE run on error spikes ends as exhausted?

When an AI SRE run on error spikes ends exhausted, raise max rounds for that Tares agent under Advanced on the agent form or as max_rounds in catalog YAML. Max rounds takes 1 to 24 and defaults to 6, or 12 once the agent uses external MCP servers. The run keeps its last text as a partial note, so you still see how far the agent got.

How it works

An exhausted run means the agent used every model round it was allowed before it reached a conclusion. Each Tares agent has a budget per run:

• max_rounds caps the model rounds. The range is 1 to 24.
• Unset, the default is 6. It is 12 once the agent uses external MCP servers, because a run that reads a diff or a file and writes back needs more than the read-and-conclude budget.
• Each call is limited to 2,048 tokens.

When the budget runs out mid-investigation, the model gets one final call with tools disabled, so it can conclude from what it has already read. If it still cannot conclude, the run ends exhausted. Tares keeps the last text as a partial note and tells you to raise max rounds.

Every run records the cap it was held to and shows rounds/max. An error-spike run that reads a wider window, checks a second entity and then writes up its note can easily need more rounds than the default 6. The fix is a larger budget, not a different trigger.

Set it up

1. Open the agent’s page and go to Runs. Find the run whose status is exhausted. Check rounds/max to see the cap it was held to, and read the partial note to see where the agent stopped.
2. Decide on a new cap. If the run stopped at 6 of 6, try 12. If it already ran at 12, go higher in steps. The maximum is 24.
3. In the console, edit the agent and open Advanced on the agent form. Raise max rounds and save.
4. Or, if you manage the agent in a catalog, set max_rounds in the catalog YAML and import it:

agents:
  - name: incident-first-look
    project: Default             # optional; the project that made it
    trigger: incident            # the trigger that wakes it
    enabled: true                # on in its project
    max_rounds: 12
    prompt: |
      You are an SRE taking the first look when an alert fires on this service. You are handed the
      correlated timeline: the firing alert plus the signals behind it and the request logs.
      Produce a tight incident note: what is failing and since when, the most likely cause tied to
      specific evidence lines, and a suggested next action. Do not speculate beyond the evidence.

5. Fire the error-spike trigger again and open Runs. The agent runs at most once per the trigger’s cooldown per entity, so wait out the cooldown first. Each agent also runs at most 50 times a day. The new run shows the higher cap in rounds/max, and a run that finishes inside it ends with a finding on the entity’s timeline.

The Build an AI SRE guide walks the full loop of trigger, run and finding if you want to check the wiring end to end.

What you get

• A run on error spikes that has enough rounds to read the timeline, look wider if it needs to, and conclude.
• A finding on the entity’s timeline, with the headline, summary and next step, in place of a partial note.
• A clear record in Runs of the cap each run was held to, so you can compare an exhausted run with the one that succeeded after the change.
• Even before the fix, the partial note from the exhausted run shows what the agent had worked out, so you do not start from nothing.

When something else fits

• If runs keep ending exhausted even at a high cap, tighten the agent’s prompt so it asks for a short, evidence-tied note. The prompt is the part of the agent you configure.
• If the runs fail rather than end exhausted, look at the error on the run and at your model provider setup. Max rounds does not affect those.
• Keep the cap as low as the job allows. A higher cap lets a run use more model calls, and the daily cap of 50 runs per agent still applies.

Related