Use a faster model
for small edits.
Jev + TypeSafe
Keep a stronger model for new pages and larger design changes. VoxelSite's optional AI Router asks TypeSafe's Jev to assess each request, then selects your writing model and the instructions it needs.
Your provider. Your model choices. Optional TypeSafe key.
Match the model to the request.
A paragraph edit can use focused instructions and the selected page's source. A site restyle needs a broader view. You choose a Faster and a Stronger model from the AI provider you already use.
Jev makes structured judgments about the work. VoxelSite maps those judgments to tested prompt recipes and model choices. Your chosen writing model generates the page.
Example routes, not a live usage report
- “Shorten this paragraph.”
- Faster model, compact editing instructions and the selected page with its shared styles and data.
- “Restyle the whole site.”
- Stronger model, restyling instructions and broader site context.
- “What does this page say?”
- A confident question route answers without changing files.
One switch. Keep editing.
Open Settings, find AI Router below AI Provider, and turn it on. Add your TypeSafe key, choose both models and save.
If Jev is unavailable, its answer is uncertain or the saved models cannot serve the request, VoxelSite reports the fallback and uses your default model. Turn the router off to use your default instructions with no TypeSafe calls.
Check what happened
Recent requests show the recorded model and instructions, including fallback to your default settings. Selected source checks protect targeted edits from overwriting concurrent changes.
Studio chat and Agent prompt execution share these settings. Direct file edits and publishing stay outside the router. Review your preview before you publish.
Before you turn it on
- You keep your generation provider
- Choose both routing models from your saved provider: Claude, OpenAI, Gemini, DeepSeek or an OpenAI-compatible endpoint. TypeSafe supplies the separate Jev classification call.
- Routing has a cost
- Smaller models and shorter prompts can reduce generation use. Each routed request also adds a TypeSafe call. We have no verified live cost-saving percentage to promise; your result depends on the work and the models you choose.
- Your request reaches TypeSafe
- VoxelSite sends the request text and a bounded record of the action, site name, page list and target paths and fingerprints. Page source and image bytes stay out of that classification call. The writing model receives its own generation context. Read the data-flow details.