> ## Documentation Index
> Fetch the complete documentation index at: https://productos.dev/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# The AI product team

> Meet the ten specialized agents behind ProductOS: what each one does, the order they work in, and how they share one project context.

ProductOS is built around a team of specialized AI agents. You describe your idea once; the team researches it, writes the PRD, designs it, codes it, and deploys it. Each agent is narrow on purpose: it has its own system prompt, its own tools, and one job it does well.

## The roster

<CardGroup cols={2}>
  <Card title="Orchestrator" icon="workflow" href="/agents/orchestrator">
    Runs the whole pipeline and routes work to every specialist. The only agent you talk to directly.
  </Card>

  <Card title="Ideation Agent" icon="lightbulb" href="/agents/ideation">
    A Socratic thinking partner that sharpens your raw idea into a testable concept.
  </Card>

  <Card title="Research Agent" icon="search" href="/agents/research">
    A market and feasibility investigator that grounds every claim in a source you can click.
  </Card>

  <Card title="PRD Agent" icon="file-text" href="/agents/prd">
    A requirements writer that works section by section, grounded in your research.
  </Card>

  <Card title="Architect Agent" icon="server" href="/agents/architect">
    An optional system design specialist for the technical half of Define.
  </Card>

  <Card title="Design Agent" icon="palette" href="/agents/design">
    A UI/UX specialist for brand, user flows, and screen specifications.
  </Card>

  <Card title="Design System Agent" icon="swatch-book" href="/agents/design-system">
    A senior visual designer for tokens, components, and real HTML previews.
  </Card>

  <Card title="Fullstack Builder" icon="code" href="/agents/fullstack-builder">
    The implementation agent that writes your actual codebase inside a live sandbox.
  </Card>

  <Card title="QA Agent" icon="bug" href="/agents/qa">
    A real-browser tester that verifies the app actually works, ending in a single verdict.
  </Card>

  <Card title="Deploy Agent" icon="rocket" href="/agents/deploy">
    Owner of the publish pipeline, from sandbox to a live URL you own.
  </Card>
</CardGroup>

## Pipeline order

Agents map to the project stages you move through in Build:

| Stage             | Agents                                       |
| ----------------- | -------------------------------------------- |
| All stages        | Orchestrator                                 |
| Stage 1: Ideate   | Ideation Agent                               |
| Stage 2: Discover | Research Agent                               |
| Stage 3: Define   | PRD Agent, plus the optional Architect Agent |
| Stage 4: Design   | Design Agent, Design System Agent            |
| Stage 5: Develop  | Fullstack Builder                            |
| On demand         | QA Agent, Deploy Agent                       |

QA and Deploy are not stages. You run them as on-demand actions: QA whenever you want the app verified, Deploy whenever you want to ship.

## One shared context

The agents do not start from scratch at each stage. Every agent reads the artifacts the previous agents produced, and the Orchestrator maintains a project wiki and cross-agent memory so nothing agreed in one stage gets forgotten in the next.

Concretely, the chain of custody looks like this: the Ideation Agent logs assumptions and open questions; the Research Agent answers them with cited findings; the PRD Agent writes requirements against those artifacts rather than inventing them; the Design and Design System Agents design the product the PRD defines; the Fullstack Builder implements it on the exact design tokens the Design System Agent locked; the QA Agent tests the critical flows listed in the PRD.

<Note>
  You never address a specialist directly. The Orchestrator is the only agent that talks to you; it delegates every research question, PRD section, design decision, and line of code to the right specialist and proposes stage transitions for your approval.
</Note>

## Model configurability

No agent is hardcoded to a specific model. The model each agent runs on is admin-configurable, and you can bring your own keys instead of using platform credits: Anthropic, OpenAI, or Google keys, or an OpenRouter gateway key. See [BYOK and models](/platform/byok-and-models).

## Where to go next

Start with the [Orchestrator](/agents/orchestrator) to understand how work gets routed, then follow the pipeline in order from [Ideation](/agents/ideation) through [Deploy](/agents/deploy). If you want the stage-by-stage product walkthrough instead of the agent-by-agent view, see the [Build surface overview](/build/overview).