> For the complete documentation index, see [llms.txt](https://docs.atlas.design/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.atlas.design/atlas-ai-studio-overview/atlas-ai-studio-overview.md).

# Atlas AI Agent

Underpinning the platform is a multi-agent system trained to help users build workflows, diagnose issues and propose improvements based on extensive 2D and 3D benchmarking not only research results, but practical out-of-distribution stress tests, and critically real-world artist usage. Every AI model we integrate into the platform gets tested against real tasks by real artists. We iterate based on what they actually ship.

Aside from allowing us to hide a majority of very specific and technical AI hyperparameters from the user (which the agent automatically sets), the agent is fundamentally a way to help someone unfamiliar with AI (like an artist) navigate and understand what these diverse models are capable of both in isolation and in combination.

**Key User Features**

* **Conversational Interface.** Describe what you need in plain language. Refine through dialogue. No prompt engineering required.
* **Context-Aware Generation.** The agent understands your project, style guidelines, and previous assets. Outputs stay consistent.
* **Real Workflow Training.** Trained on professional game art pipelines, knows the difference between concept and production as well as various art styles and direction.

**Key Technical Features**

* **Context engineering.** The agent decides what goes into the context window. You don't manually copy-paste code files and error logs. The app does the retrieval, embedding, and curation. This is a ton of hidden work.
* **Multi-call orchestration.** Under the hood, there are embedding models for your files, chat models for reasoning, models that apply nodes. The user sees one experience. The app runs a whole orchestra.
* **Application-specific GUI.** Text is hard to audit. Seeing nodes and connections uses your visual system, which is way faster than reading, and it gives the artist way more control.

<br>

## When to lean on the agent

The agent is most valuable in three situations:

* **First-time workflow construction.** Describe the production target in plain language (e.g., "Generate game-ready props from concept images at 5k polycount with PBR materials for Unreal Engine") and let the agent propose a starting workflow. Iterate from there.
* **Debugging unexpected output.** If a workflow is producing low-quality results, the agent can analyze the chain and propose specific node changes (different backend, different hyperparameter, additional pre or post-processing step). This is faster than trial-and-error on the user's part.
* **Discovering new capabilities.** The platform integrates a wide range of models and node types. The agent surfaces relevant capabilities the user might not know about, especially when crossing between domains (e.g., from image generation into 3D, or from a 3D mesh into animation).

For experienced users running a known workflow at scale, the agent is less essential. The biggest leverage is in early exploration and in any situation where the production target is shifting.

## Common pitfalls

* **Vague prompts produce vague workflows.** "Make a 3D model" is a weaker instruction than "Generate a 5,000-polycount stylized prop from this image with PBR materials and a tri-only topology for Unreal Engine." The richer the production target you communicate, the more precise the agent's output.
* **Skipping the agent and wiring nodes manually too early.** New users sometimes treat the agent as a beginner crutch and try to wire every node by hand. The agent is faster for almost all workflow construction, including for experienced users. Lean on it.
* **Ignoring the agent's debugging suggestions.** When a workflow produces unexpected output, the agent's proposed changes are usually based on benchmark patterns across many real production runs. Try its suggestions before reverting to manual fiddling.

## Related pages

* [Getting Started](/atlas-ai-studio-overview/getting-started.md) — broader orientation to the platform
* [3D & Gaming Focus](/atlas-ai-studio-overview/3d-and-gaming-focus.md) — the production-grade 3D context the agent operates inside
* [Node Index](/atlas-ai-studio-overview/node-index.md) — the full set of building blocks the agent can compose
* [Testimonials](/atlas-ai-studio-overview/testimonials.md) — customer quotes consistently highlight the agent as the differentiator

## Frequently asked questions

**Is the Atlas AI Agent a single model or a system?**

It is a multi-agent system, not a single model. Under the hood, different components handle context retrieval, reasoning over workflow construction, hyperparameter selection, and execution against the node graph. The user sees one conversational interface; the system runs multiple specialized agents.

**Can the agent build any workflow, or are there limits?**

The agent operates over the full set of nodes available in Atlas. If a node exists in the platform, the agent can incorporate it into a workflow. For capabilities not yet exposed as nodes, the agent cannot improvise.

**Does the agent require prompt engineering skills?**

No. The agent is built for artists and production teams, not prompt engineers. Clear descriptions of the production target (engine, polycount budget, texture resolution, style direction) are far more useful than carefully crafted prompts.

**How does the agent stay current as new models are added to Atlas?**

The agent's model coverage updates as new AI models are integrated into the platform. Every new model gets benchmarked against real tasks by real artists before it is exposed through the agent, so the agent's suggestions reflect production performance, not just research claims.

**Can the agent be used inside an exported API workflow, or is it only available in the web UI?**

The agent is the workflow-construction layer; once a workflow is built and tested, the exported API runs the workflow itself, not the agent. Workflows produced by the agent are first-class Atlas workflows and can be exported as APIs like any other.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.atlas.design/atlas-ai-studio-overview/atlas-ai-studio-overview.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.