Reference
Models
PromptFloe routes your requests to the best model for each stage of the pipeline. You can override the model for advanced use cases — most users don't need to.
#Default routing
Each pipeline stage uses a model tuned for its job:
| Field | Type | Description |
|---|---|---|
| Planner | reasoning | Long-context reasoning model. Plans the file manifest. |
| Files agent | fast structured | JSON-mode-tuned model. Emits the file tree. |
| Code agent | code-specialized | Per-file generation. Multiple instances in parallel. |
| Reviewer | reasoning | Cross-file checks for missing imports, unused symbols. |
| Critique skill | general | Long-form markdown reasoning. |
| Augmenter | code-specialized | Same model class as code agent — produces strict JSON patches. |
#Per-call overrides
Pass modelOverride on the generate or run call to force a specific model. Available aliases:
| Field | Type | Description |
|---|---|---|
| fast | alias | Smaller, cheaper model. Lower quality, faster turnaround. |
| balanced | alias | Default. Best price/quality tradeoff. |
| quality | alias | Most capable model. Higher cost, slower. |
| reasoning | alias | Long-thinking model for complex planning. |
await client.apps.generate({
prompt: '...',
modelOverride: 'quality', // use the highest-tier model
});#Enterprise: bring your own model
Enterprise plans can wire custom inference endpoints — your own fine-tuned models, on-prem inference clusters, or alternate providers — through a proxy contract. Contact sales for setup.
#Where to go next
PromptFloe developer docs