Agent runs

It stopped part way through

A single turn has a limit on how many steps the agent takes. On a long build it may stop and describe its next step. Reply continue and it picks up where it left off.

If a run ends with no reply at all, send “continue from where you stopped”. Long code steps can time out; asking it to split the work into smaller steps helps.

It ignored a file you uploaded

The agent writes its plan before it looks at files, so a file it isn’t told about may not make the plan. Name each file in your message, in the order to read it: “Read plan.md, then style.md, then build the report.” Standing instructions go in the project’s AGENTS.md (see Projects and standing instructions).

You can’t find the output

Open the File Manager from the message box. Finished deliverables are saved in the project’s Final Reports folder, and working files such as CSVs and plans are in the project’s files. Ask “Where did you save the report?” if it isn’t there. To share it, see Share it.

The answer used a workaround

When Dimensions it needs are empty, the agent may answer from the raw text and say so. For a firmer answer, run a backfill from the Backfill tab and ask again once it has finished, typically within the hour. See No Dimensions yet.

“Today” is wrong

The agent’s clock runs on UTC, which is already tomorrow in the Americas in the evening. Give explicit dates for “this week” or “last month” questions asked late in the day. The platform’s own date fields are in US Pacific time; see A date is a day off.

It can’t save a Dimension

The agent can draft a Dimension in any chat, but it saves prompts only when Prompt Designer is on. Turn it on in the agent’s settings and ask again. See Designing Dimensions with the agent.

It can’t help with your account

The agent works on your data, not your account. For login, billing, users or anything else, email [email protected].