> 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/release-notes/release-0.45.0.md).

# Release 0.45.0

Previous Version: **v0.44.0**

Discover workflow-powered apps in Atlas Labs, preview a node's inputs and outputs before adding it, and compare historical execution times alongside model costs. Connected-tool setup is clearer, payment warnings arrive earlier, and the AI Assistant can consult relevant application guides while helping you.

## Discover workflow-powered apps in Atlas Labs

Browse **Atlas Labs**, a new catalog of apps built on Atlas workflows. Cards show descriptions, cover images, media types, and workflow counts, with a larger card for the featured Lab.

On desktop, open a Lab inside Atlas to see the app, its workflow sequence, and run availability together. Labs that support running inside Atlas use the selected workspace's credits and your existing member limits. Smaller screens offer **Open in a new tab**.

The catalog appears when Labs have been configured. Some Labs are prepared examples and show runs as unavailable until their hosted app supports running inside Atlas.

![Browse workflow-powered apps, media types, and workflow counts in the Atlas Labs catalog.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fr5LP0jbF6ucQVaGBOL6A%2Flabs-catalog.jpg?alt=media)

![Open a Lab inside Atlas and check its workflow sequence and run availability.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FrAqBTjkjO7RG7BCbtmi4%2Flab-desktop.jpg?alt=media)

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F377s17enjae3cxcEnJx5%2F20261022-1243-34.8418125.mp4?alt=media&token=8ec272b8-83cb-4686-8418-389d94b8c92f>" %}
Outfit Builder App
{% endembed %}

## Run shared exported APIs from your own workspace

Exported APIs that Atlas staff make available across workspaces can now be called with your own workspace API key, through both synchronous and asynchronous execution.

Runs use your workspace's credits, limits, uploaded inputs, and output assets. Execution status and results stay scoped to the workspace that started the run. Existing execution status remains readable if staff later stop accepting new runs across workspaces. Access to run an API does not grant permission to manage it.

## Preview a node's inputs and outputs before adding it

Hover over or keyboard-focus a node in the **Add node** menu to see its name, variant tag, description, and color-coded input and output chips. The chips distinguish individual values from lists, helping you check what a node takes and produces before adding it to your graph.

![The Image to 3D preview shows its FAST variant tag, description, and IMG-to-3D input and output chips.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FgrqGarK6h7NR9eKa6gZP%2Fnode-preview-variant.jpg?alt=media)

## Compare historical execution times alongside model costs

Node model pickers now show historical execution durations beside credit-cost indicators where data is available. Models with sufficient history show a range; smaller samples show an approximate average. A tooltip explains the measurement.

Durations summarize successful, uncached executions across settings and exclude Atlas queue and billing time. Use them to compare options; they are not predictions for a particular request. Options without published history show no duration.

The screenshots below use illustrative timing data to show the interface. They are UI examples, not measured model benchmarks.

![Illustrative timing data: a model picker shows a range and an approximate average beside credit-cost indicators. Actual duration labels require published history.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FCw0OLgQpF2pHdC3A6cwx%2Fmodel-execution-durations.jpg?alt=media)

## Manage MCP connections from a clearer page and approval screen

The full-width **MCP** page puts **Connected tools** first, with recognizable tool logos and indicators for live connections. Last-call information refreshes automatically every **10 seconds** while the page is visible. Disconnecting opens an in-app confirmation explaining what happens to access and work already running.

Tool approval now opens a standalone **Access request** screen showing the signed-in account, selected workspace, and credit cap per agent run. Clearer pending, expired, canceled, and approved states help you follow the request. Approvals made in a workspace name that workspace, and a create-workspace link is available when needed.

![Connected tools appear first, with recognizable logos, live-connection indicators, and automatically refreshed last-call information.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FU7pRv8sh09bOEYWzUMuo%2Fmcp-connected-tools.jpg?alt=media)

![The standalone Access request presents the selected workspace and credit cap before approving a connected tool. Personal account details are omitted.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fp6LW5PltFWGYQ9NYYI9c%2Fmcp-access-request.jpg?alt=media)

![The in-app disconnect confirmation explains what happens to access and work already running.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fv0612RP92L0zBAUavWCF%2Fmcp-disconnect-confirmation.jpg?alt=media)

## See payment problems from the first failed attempt

Payment warnings now appear from the first failed subscription or on-demand payment, before the workspace reaches the existing payment-lock threshold. The warning identifies the charge and amount, with Stripe's next retry date when available.

Check payment status in the dashboard or desktop editor banner, **Plan & credits**, and the credits popover. Workspace owners and admins can choose **Update payment method**; other members receive guidance to contact a workspace owner.

While Stripe retries a past-due subscription payment, the plan card retains its plan information, shows **Payment failed**, and displays the next retry date when available. The credits popover prioritizes payment recovery over upgrade offers.

An early warning alone does not block runs. The restricted state remains separate and follows the existing payment-lock policy.

![The credits popover prioritizes subscription payment recovery over upgrade offers.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FpMO7KoPmiYvW4rN8rqWt%2Fpayment-credits-popover.jpg?alt=media)

![The credits popover identifies the failed on-demand payment and offers payment recovery.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fu6AD8Nry3lpHaofllquR%2Fpayment-on-demand-popover.jpg?alt=media)

## Let the AI Assistant consult relevant application guides

When answering questions about navigation, settings, or controls, the project **AI Assistant** can now retrieve relevant application-guide topics on demand. Its activity shows **Reading application guide** while doing so, making this part of its existing application help visible as it works.

![The completed AI Assistant conversation retains Reading application guide activity for the relevant topics.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FzaOXNoG5fZRw1JA2xYs6%2Fassistant-guide-activity.jpg?alt=media)

## Upload attachments to your workspace or project through MCP

Connected AI tools can now send attachments into Atlas with **Upload workspace asset** and **Upload project asset**. In ChatGPT, attach a file to the conversation; other clients can provide an accessible public HTTPS download URL.

Workspace uploads add reusable assets for all workspace members. Project uploads add a matching input node to an editable canvas, ready for your next workflow run, while keeping the project's existing privacy and sharing settings.

Project uploads support images, meshes, audio, video, PDFs, and Blender scenes. Files can be up to **100 MiB**, within the project's existing total upload limit.

![An attachment uploaded through MCP appears in the workspace's reusable assets.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2Fxa3Q7jYU03ZJtOZr0CcU%2Fmcp-workspace-upload.jpg?alt=media)

![An image uploaded through MCP is ready to use in a resize workflow.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FyT4XggYibxkxtTAzfWgR%2Fmcp-image-workflow.jpg?alt=media)

## See the final charged credits for a generated asset

Open a generated asset's details to see **Credits: X** for the node run that produced it, after applicable refunds. Outputs and intermediate assets from the same run show that run's charge; upstream runs have their own costs.

Cached assets keep their original producing-run charge. A recorded **0** means no credits were charged. Assets without a recorded charge omit this field.

![The producing text-to-image node completes with a final charge of 1 credit.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FDICw5sP8YSNsktUrkqai%2Fgenerated-image-workflow.jpg?alt=media)

![This generated image records the producing node run's final charge of 1 credit.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FnF4KpQSoSvzVYVfcTNFM%2Fasset-final-credits-1.jpg?alt=media)

## Find nodes under clearer task headings

The **Add node** menu puts AI creation groups first and organizes longer lists with sticky task headings. Browse sections such as **Generate**, **Look & style**, **Framing & size**, **Texture**, **Geometry**, **Edit**, and **Convert** to find the operation you need.

List management nodes now sit under **Utility Nodes**. Procedural image building and character-sheet splitting sit under **Images**.

![Sticky task headings organize a longer node list into operations you can scan.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FcSyVjRO4XuDBfu2so6qG%2Fmenu-image-task-headings-full-app.jpg?alt=media)

## Use planning-based upfront credit reservations for image runs

Quality **Text to image**, **Text + image to image**, and **Text + image list to image** can now use a planning-based upfront reservation from the selected model settings, references, and required processing, including when a trained estimate is not yet available.

The node’s **Up to** amount previews the run’s charge ceiling; unused reserved credits are returned. This changes the reservation policy without changing output quality or provider prices.

![Quality Text to image previews an Up to 6-credit reservation for the selected Nano Banana 2.1 1K configuration.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FmQr1xSbJUjY8ZDGSjbNE%2Fimage-reservation-text.jpg?alt=media)

![Text + image to image previews an Up to 13-credit reservation with one reference, Grok Imagine 2.0, and 2K Medium settings. The displayed image is the previous result.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FZMGL9upQLr8BTtuMr9Pl%2Fimage-reservation-reference.jpg?alt=media)

## Choose Google Gemini 4 Argon for language and agent tasks

Google Gemini 4 Argon is available in the applicable language-model and agent-node model selectors. Choose Low, Medium, or High reasoning; Medium is the default. Existing text, supported media, structured-output, and tool-use controls follow the selected node.

Atlas agents using this model retain its native conversation context across tool exchanges and saved conversations. Model support is built into existing nodes and agent flows.

<figure><img src="https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FXoYCZVQVutfbW8AWdUNa%2Fimage.png?alt=media&amp;token=5ba7dc30-fa66-481e-b5a2-a228f573b64f" alt=""><figcaption></figcaption></figure>

## Set a credit budget for Procedural Image Builder

The experimental Procedural Image Builder has an advanced Maximum Credits input. Set it manually or connect a numeric input to cap one execution. It defaults to 1,000 credits and accepts values from 1 to 100,000.

The node checks whether its next model or worker step fits the remaining budget before starting that work. Unused reserved credits are refunded. If the run stops after generating images, those images can be returned with a warning and are not cached; stopping before any image is generated can produce an error instead. A larger budget permits more work but does not guarantee completion.

![The advanced Maximum Credits control with the default 1,000-credit budget in a completed procedural-image example.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2F2YA7yPo7wL0qTdWPStIq%2Fr045-r5-procedural-credit-budget.jpeg?alt=media)

## Use planning reservations in more image tools

Planning-based up-front credit reservations extend beyond the Quality image nodes to supported modes of Fast Text to image, image LoRA control and inference, Outpainting, Upscale, Material Generation, text/image conversion to SVG, Image Layers, background removal, depth estimation, and character-sheet splitting.

The quote accounts for selected options, supplied references, and required internal processing. Some modes retain their full maximum. Output quality and execution limits remain governed by the selected tool; a lower final charge is not guaranteed. Procedural Image Builder uses its separate explicit credit budget described above.

## Connect a dropped wire to the first numbered input

Drop a wire on the canvas and choose a new node with numbered inputs to connect it to the first compatible numbered input automatically. Nodes with distinct named input choices still ask which input to use.

{% embed url="<https://files.gitbook.com/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FkhguSJZRqAOERy3zwB8w%2F20261026-0645-02.5392779.mp4?alt=media&token=126a226c-4dcb-470a-ad57-16c35e6545b3>" %}

## Recover saved files from unsuccessful Blender runs

Unsuccessful Blender runs can retain completed recovery files, such as a Scene, Conversation, or applicable mesh/image artifact. Use Saved partial files · inspect or download beside the failure to recover work that finished before an error, cancellation, or lost session.

Recovery depends on which files were actually completed and retained. The run keeps its unsuccessful status, with recovery files separate from successful history and cached outputs. Files are available only when work was completed and retained before the failure.

![An unsuccessful Blender run retains Scene and Conversation recovery files for inspection or download.](https://3654894688-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FR7boiMixMhR4q36Ns33Y%2Fuploads%2FNDmZjiDswGX8mHz3Zziq%2Fr045-r5-blender-recovery.jpeg?alt=media)


---

# 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 by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.atlas.design/atlas-ai-studio-overview/release-notes/release-0.45.0.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

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.
