Start and run a plan

Once a plan is created, you can trigger it manually or through conversation. This article covers starting a run, following its progress, and what to do when a step fails.

Starting a plan

From the Plans tab

Select the plan from the Plans tab and start it directly.

From a bookmark

Bookmark a plan to have it appear in the sidebar for the assistant it belongs to, ready to run without navigating to the Plans tab first.

From a conversation

Ask the assistant to run a specific plan in chat.

Providing required files

If the plan has expected inputs, upload the required files before it starts. If you upload a file in the chat and then ask for the plan to run, that file is automatically made available to every step that needs it.

How a step runs

Each step runs as an isolated task with its own context — steps execute sequentially, and the next one can’t start until the previous one completes.

What each step receives

• The step’s own instruction.
• The assistant’s prompt and knowledge.
• Access to every tool enabled for the assistant.
• Any relevant attachments or files.

How a step reaches earlier results

Steps don’t automatically receive the output of earlier steps. Instead, the assistant fetches specific outputs or tool calls from previous steps as needed. Step five, for example, can pull data generated specifically in step two, rather than everything every step has produced so far.

Following a run

Each execution shows real-time status: running, completed, or failed overall, and each step marked as completed, in progress, or failed individually.

Cancelling a run

Cancel a run in progress at any time from its execution view.

Re-running from a step

If a run didn’t go the way you wanted, re-run it from a specific step rather than the whole plan over again — that step and everything after it runs again, replacing their previous results.

Generated files and artefacts

Open an execution’s details to see what each step produced — generated files, created artefacts and media, and which tools were called along the way.

Understanding failures

• Step-level failure — the assistant determined it couldn’t complete that specific task. Depending on the plan’s logic, remaining steps may still attempt to run.
• System-level failure — a platform issue, such as a rate limit or a provider outage, rather than something the assistant decided. These are raised programmatically rather than by the assistant itself.

Open the execution’s details for the technical error information behind either kind of failure.