Aetherion AI

Operating Workspace

AI Builder
Guide6 min~269 words

Choose OpenAI, Gemini or Claude

Understand price-plan access, personal defaults and per-operation model selection in Talk, Build and Benchmark.

How Aetherion chooses a model

PrioritySourceCondition
1Manual dropdownThe current price plan allows manual selection and includes the exact model.
2User Settings defaultThe plan allows personal defaults.
3Price-plan defaultConfigured by Admin separately for Talk, Build and Benchmark.
4Platform defaultUsed only when it is also allowed by the price plan.

A dropdown never bypasses the plan

The server validates provider and model on every paid request. A stale or manipulated selection is rejected instead of silently routing to a different or more expensive model.

Set defaults or choose for one operation

01

Set personal defaults

Open Account Center → Settings → Default LLM and choose separate models for Talk, Build and Benchmark when your plan permits it.

02

Choose in the composer

Use the model dropdown beside Build / Talk to override the default for the next operation.

03

Confirm in Build Setup

The prebuild form repeats the primary Build model beside layout, typography and visual direction.

04

Choose in Benchmark

Select an allowed web-grounded model before reviewing the estimate and exact question fingerprint.

What the Build model controls

The selected LLM is the primary reasoning model for Talk or the Builder plan, including length recovery and JSON repair. Aetherion's provider-neutral Build Engine still performs repository evidence collection, action validation, virtual file simulation, security checks, versioning and controlled file writes. Changing the model never grants direct access to files or bypasses those guards.

Model tokens and provider web searches are recorded under the actual provider and model in Money Flow. Different models can therefore produce different estimates and credit settlement even for the same prompt.

Need project-specific help?

Open the private Help Center or create a support ticket with the affected project, route and request ID.