> For the complete documentation index, see [llms.txt](https://alludium.gitbook.io/alludium-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://alludium.gitbook.io/alludium-docs/12.-workspace-overview/12.5-ai-model-resolution.md).

# AI Model Resolution

Workspace Owners and Admins can control which AI model Alludium uses when an agent or task starts an assistant pass.

Most users should leave model selection on the inherited default. Custom model choices are available for workflows that need a specific cost, capability, context window, output size, or reasoning behaviour.

## Resolution hierarchy

When Alludium starts an agent pass, it resolves the model in this order:

1. **Task runtime override** - a model deliberately selected on a specific task run.
2. **Task type default** - a model set on the task definition.
3. **Agent custom model** - a model deliberately selected for the agent.
4. **Project default** - a model set on the specific project.
5. **Project setup version default** - a model set on the project type version that the project uses.
6. **Workspace default** - the workspace-level default model.
7. **Platform default** - the platform-managed default model.

The first resolvable custom/default entry wins. Missing context is skipped.

## Why this order

The hierarchy separates **explicit overrides** from **contextual defaults**.

Task and agent selections are treated as explicit overrides because a user has deliberately customized that execution surface. Project and project setup defaults are contextual: they guide work inside a project or workflow type, but they should not override a task or agent that has already been customized.

Workspace and platform defaults are the final baseline. They keep ordinary agents and tasks working without requiring every workflow to choose a model.

## Context examples

| Execution context                      | Resolution path                                                                                                                                               |
| -------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| My Agents chat with no task or project | Agent custom model -> workspace default -> platform default                                                                                                   |
| Inbox task with no project             | Task runtime override -> task type default -> agent custom model -> workspace default -> platform default                                                     |
| Project task                           | Task runtime override -> task type default -> agent custom model -> project default -> project setup version default -> workspace default -> platform default |
| Project chat with no task              | Agent custom model -> project default -> project setup version default -> workspace default -> platform default                                               |

## Customizing defaults

Use the shared model selector wherever model configuration is available. It shows the inherited model first and only opens the full picker when you choose to customize.

Changing reasoning or temperature also creates a custom override. If you customize the inherited default model only to adjust reasoning or temperature, Alludium locks in that model as part of the override. Later changes to the upstream default will not change that customized surface until you reset it.

Use **Reset to default** to return a task, agent, project, or project setup version to inheritance.

## Model availability

The model picker only shows models that the platform can use with an available infrastructure provider connection. If a model provider or infrastructure provider is not available in the workspace/platform environment, its models are hidden from selection.

Model cards can be sorted by price, release/update date, context window, and output limit where metadata is available. Exact prices are not shown on the cards.

## Audit and billing metadata

For each runtime assistant pass, Alludium records the resolved model source and selected model metadata for operational clarity. This includes the source that won the hierarchy, the resolved model identifier, model provider, infrastructure provider, and internal model ID where available.

This metadata supports task activity/audit history, runtime diagnostics, and billing attribution. It is especially useful when a task inherits through several levels and the visible surface does not have a direct model override.

## Administration guidance

Use the broadest default that matches the policy:

* Set the **workspace default** for the normal model/cost profile of the workspace.
* Set a **project setup version default** when every project of that workflow type should inherit the same model.
* Set a **project default** when one project needs different economics or capability from its setup.
* Set an **agent custom model** when that agent's purpose requires a specific model everywhere it is used.
* Set a **task type default** when a repeatable task needs a specific model independent of project context.
* Set a **task runtime override** only for a one-off run or active task-specific exception.

Do not use model defaults as access policy. If the workspace needs hard caps or compliance restrictions, those should be managed as platform policy rather than by relying on default inheritance.


---

# 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:

```
GET https://alludium.gitbook.io/alludium-docs/12.-workspace-overview/12.5-ai-model-resolution.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
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.