Set the task
Give the agent a question, Project context, and a clear result to produce.
AI
Use AI inside a Project.
Agents can read Project Resources, search when allowed, write code, build Tools, and run experiments. A Run records the work and its cost.
AI works inside the Project. Its sources, actions, corrections, and results remain available for review.
A Run
A Run starts with your choices.
Set the task; ResearchTogether resolves your saved model and payer, then pins the mode’s bounded limit before the agent begins.
Choose one default model
Set the model and payer once in AI Settings. New work applies the bounded limit for its mode automatically.
Follow the Run
Inspect actions, sources, outputs, and failures while the agent works.
Review the result
Check the evidence, method, code, and conclusions. Revise or rerun when needed.
Publish what holds up
A person decides what becomes part of a durable Version.
Agent work
What agents can do.
Agents work with the Resources and capabilities available inside the Project.
Research a question
Read Project Resources, compare evidence, and search outside sources when allowed.
Write and run code
Implement an analysis, run an experiment, and report failures with the result.
Build a Tool
Create a purpose-specific Tool when the same method needs to be used again.
Boundaries
What does not happen automatically.
Using an agent does not remove the choices that belong to people.
AI is not required
You can read, create Projects, collaborate, and publish without invoking a model.
The model and payer do not switch silently
If a credential, sponsorship, permission, model, or provider is unavailable, the request stops.
A Run does not hide its trail
The model, activity, sources, outputs, failures, usage, and cost stay with the Run.
AI does not publish for you
People set the direction, review the work, and choose what enters a Version.
Start a Run