Threads
Follow your agent’s work, refine the result, and keep the job moving.
A thread is where an agent handles a job. It brings together the directions, work in progress, questions, and results so you can follow what happened and build on it.
Agents can start work from your message, a schedule, a webhook, or a request from another agent. Depending on the setup, the request starts a new thread or continues an existing one.
A run is a period of work by the agent. A thread holds the job across those runs. The agent might prepare a renewal brief in one run, then revise it in another after you supply updated pricing.
See your agent at work
Follow progress, tools, questions, and outputs.
Progress and reasoning
Follow status updates, visible reasoning summaries, and the steps the agent takes toward the result.Tool activity
See searches, browser actions, calculations, and other work. Expand activity to inspect the details available for that step.Questions and approvals
Answer questions and review approvals when the agent asks. Your answer helps it continue the job.Outputs
Open documents, tables, webpages, and files as the agent creates them. Review the result alongside the thread and ask for changes.
The work can continue while you’re away
Closing the browser does not stop an active run. Return to the thread to see the result or respond if the agent needs your input. Runs still have time limits and can stop when they encounter a problem.
For settings and run controls, see Thread settings and controls.
See what your agent delivers
From a quick answer to a finished webpage or an update in your tools.
Read the summary
The agent can answer a question, explain a finding, or recommend a next step directly in the thread. Its reply can summarize the work and link to sources and finished outputs, so you can get the main points before opening the details.
Open the finished work beside the thread
Select an output card to open it in the canvas. Review a document, compare a table, explore a webpage, or preview supported files and media while keeping the conversation alongside it.
Ask for a format that fits the job: a webpage to present research, a spreadsheet for analysis, a document to explain a recommendation, or images, video, and audio. Saved outputs are also available in the Library.
Get work done where it belongs
A result can be an action: an updated customer record, a new task, or a document written to a connected service. The thread shows the activity and confirmation, so you can follow what changed and open the result where a link is available.
What the agent can change depends on its integrations, access, and approval settings.
See what your agent produces for more about outputs and saved work.
Guide the work
Give feedback, correct misunderstandings, and refine the result.
Help the agent understand the job
Add a source, clarify the goal, or explain what matters most. Keep directions in the same thread when they belong to the current job. See how sources contribute to context.
Check the directionPrompt
Before going further, tell me what outcome you’re working toward and the assumptions you’re making. Identify anything unclear that could materially change the result, and ask me about it. Then suggest the most useful next step.
Be specific about what should change
Tell the agent what to keep and what to improve. You can revise the content, format, or approach without starting the job over.
Make the result more usefulPrompt
Review this against my original request. Fix anything incomplete or unsupported, cut repetition, and make the main finding or recommendation easy to spot. Keep the useful detail, distinguish facts from assumptions, and tell me what still needs my input.
Keep useful feedback for future jobs
Ask the agent to save a lasting preference as a memory or a repeatable method as a skill. Review suggested learnings where your learning settings require it.
Save the approachPrompt
Review the feedback I’ve given in this thread. Identify preferences worth remembering and methods worth saving as a skill. Separate those from one-off directions and facts specific to this job. Propose what to keep, explain when it would help, and ask me before saving it.
Make a useful workflow repeatable
Once you have refined how the job should be done, create an agent you can use for similar work. Add a schedule when you want it to run regularly.
Share results
Share, send, publish, or deliver the result where it’s needed.
Keep the result in the thread, share a link, or have the agent deliver it through a connected service. The thread keeps the record of what was created and delivered.
Send to Slack
Have the agent post the brief and its key findings to the channel where your team works.
Email it
Ask the agent to send the result through your connected email account, with a summary for the recipient.
Write to another service
Update a customer record, add a document, or create a task in a connected tool.
Delivery depends on the connected tools and approval settings available to the agent. See sharing and delivery options.
Explore threads
Learn more about the workspace, context, and controls behind the work.
What’s in a thread
Explore the tools, computer, notes, and outputs behind the work.
Thread context
Understand what the model can work with and how context changes.
Thread settings
Adjust settings for this job and review suggested learnings.
Turn a thread into an agent
Reuse a workflow that works for future jobs.