Platform · Model intelligence
Models will change. Your intelligence layer shouldn't have to.
Different models are better suited to different tasks, budgets, privacy requirements and reasoning workloads. QuerySafe's longer-term platform direction is model-aware rather than defined by one provider.
Understand the task
What is being asked, and how hard is it?
Understand the constraints
Cost, latency, privacy and policy.
Use the appropriate intelligence
The model that fits both.
What shapes the choice
Task, budget, privacy and reasoning, decided per request.
An illustration of how a model-aware layer routes four everyday requests.
| Request | Main constraint | Routed to | Why |
|---|---|---|---|
| Summarize yesterday's sales | Speed · low cost | Fast | Short, factual, well-structured input |
| Explain a drop in regional revenue | Accuracy | Balanced | Multi-step analysis over query results |
| Draft a board note on supplier risk | Reasoning depth | Deep reasoning | Long-form synthesis with trade-offs |
| Answer from HR policy documents | Privacy policy | Private / local | Sensitive text stays inside the boundary |
Illustrative. Private and local routing is part of our longer-term direction and QuerySafe Labs.
Where this shows up
QuerySafe Compass applies the same thinking to everyday AI tools. QuerySafe Labs explores private model routing: choosing between local and cloud models based on task and policy.
FAQ
Model questions
Which AI model does QuerySafe use?
QuerySafe's platform direction is model-aware rather than tied to one provider, so that the layer your teams use doesn't have to change when models do.
What is AI model routing?
Sending each task to the model best suited to it. A short factual summary may need a fast, low-cost model, while a multi-step analysis benefits from stronger reasoning.
Is routing to private or local models available?
Not yet. Private and local routing is part of our longer-term direction and QuerySafe Labs. The routing examples on this page are illustrative.
How does this relate to QuerySafe Compass?
Compass applies the same idea for individuals: it improves prompts and recommends the right model for each query. Compass is live now.
Why not just pick the best model?
Because the best model for a hard analysis is rarely the best choice for a quick summary. It costs more and responds more slowly. A model-aware layer can match the task to the model instead.