Skip to content
EndueEndue
Docs

Runs

A run is one turn of an agent actually working: it reads its context, decides, calls a tool, reads the result, and repeats until it has an answer or needs you.

You do not start a run directly — sending a message starts one. What this page is for is reading one: knowing what the agent did, why it stopped, and what to do about it.

Each step appears in the conversation as it happens:

  • A tool call, with the arguments the agent chose. This is the most informative thing on the screen. An agent that searched for the wrong thing tells you more than a bad final answer does.
  • The result that came back, including errors. A failed call is not the end of a run — the agent sees the error and can try something else.
  • The agent’s text, streamed as it is produced.

If the final answer is wrong, scroll back through the tool calls. The mistake is almost always visible in one of them, and what to change follows from which one.

It stopped becauseYou seeWhat to do
It finishedA final answerNothing
It needs information only you haveA question — options, a text field, or a set of preferencesAnswer it; the run resumes
It wants to do something irreversibleAn approval request showing the exact actionApprove or reject
It wants a skill it does not haveA request to grant the skillGrant or decline
It hit a step limitThe run ends without a final answerAsk it to continue, or narrow the task
It failedAn errorRetry, or fix what it was blocked on

A paused run is holding its place, not starting over. Answering resumes it with everything it had already worked out.

Runs execute on Endue’s servers, not in your browser tab. Closing the app does not stop the work.

Start a long task, close the tab, come back later — the run kept going and the result is in the thread. The conversation list marks threads that finished while you were away, so you do not have to remember which ones to check. This is also what makes routines possible: a scheduled run does not need anyone to be watching.

  • A run has a step limit. A task that needs more steps than the limit allows ends without a final answer — split it, or narrow it.
  • A run keeps the configuration it started with: prompt revision, model, and bound tools. Changes made mid-run apply to the next one.
  • You cannot edit a tool call the agent already made. You can steer what it does next, or reject an action at the approval gate.
  • There is no replay. Re-running a task starts a fresh run, and agents are not deterministic — the same request can take a different route.