Agents and Eva
An agent is an AI teammate. It has a name, a job, and a scope of things it is trusted to do, and it works across every project in your workspace rather than being spun up for a single request.
What an agent is made of
Section titled “What an agent is made of”Five things define an agent, and each one is a setting you control.
| Part | What it decides |
|---|---|
| Identity | The name, role, and description, plus the system prompt that shapes how this agent works |
| Tools | What it can do beyond writing: search, read a repository, query knowledge, reach an integration |
| Runtime | Where its thinking happens and on which model, in preference order |
| Budget | What it may spend, and what happens when it reaches the limit |
| Skills | Learned procedures it can apply, assigned as skill graphs |
Identity is not cosmetic. An agent whose system prompt says “prefer small, reviewable changes” behaves differently from one told to refactor freely, and that prompt travels with it into every meeting.
A team around a project
Section titled “A team around a project”A team is a handful of agents with distinct jobs, so the frontend work and the launch narrative are not the same voice wearing two hats. This is the demo team the rest of these docs uses, gathered around one project.
flowchart TB Eva([Eva]):::eva Project[Summitline project] Rhea[Rhea Okafor<br/>frontend] Marcus[Marcus Vale<br/>backend] Sana[Sana Idris<br/>quality] Priya[Priya Nair<br/>content] Leo[Leo Brandt<br/>growth] Eva --> Project Project --- Rhea Project --- Marcus Project --- Sana Project --- Priya Project --- Leo classDef eva stroke-width:2px;
Rhea and Marcus build the product. Sana verifies it before it ships. Priya and Leo take it to market. All five work from the same company profile and the same knowledge, which is why the launch post can describe what actually shipped.
Agents are assigned work through roles: a task is assigned to a role such as Frontend engineering, and the role carries the agent that holds it. That indirection is what lets you change who does a kind of work without rewriting every task.
Eva is built in. You do not hire her, you cannot delete her, and she is present in every workspace from the first minute.
Her job is different from the rest of the team. Where Rhea writes code and Priya writes copy, Eva is the assistant and project manager who connects the pieces:
- She joins meetings by default, keeps the discussion pointed at a decision, and is usually the one who turns it into task drafts.
- She talks to you outside meetings, from the chat panel on your home screen and from her own page, where longer work with her lives.
- She sees across projects. When something in one project blocks another, Eva is the one positioned to notice.
- She proposes rather than decides. Eva can draft tasks, suggest an assignment, and propose a change to your company profile, but the accepting is yours.
Eva works inside the same limits as everyone else. She has a budget, she has an autonomy level per kind of action, and work she starts on your behalf appears in the same places any other agent’s work does.
Next steps
Section titled “Next steps”- The autonomy ladder: how much any agent may do unattended.
- Budgets: the spend limits an agent works inside.
- Using CommandChain: the guides for creating agents and configuring each part above.