Answers from the documentation; checking which model is on duty
Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Privacy-first routing - Template - OpenSmartRoute
TemplateFeaturedv1.0.0
Privacy-first routing
Anything with personal data is pinned to an on-prem model; everything else uses the cloud tiers.
For teams under a data-residency or PII policy: the router detects personal data and pins those requests to a self-hosted model that never leaves the boundary, while ordinary traffic still uses the faster cloud tiers. Swap llm-onprem for your own in-network deployment.
Use it
Copy one of these into your project. Installing also returns the manifest and these snippets.
A declarative stack: models, rules and settings in one file. `osr stack apply` turns it into a running router.
privacy-first.stack.jsonjson
{
"osr": "1",
"kind": "stack",
"name": "privacy-first",
"description": "Personal data stays on-prem; everything else uses the cloud tiers.",
"targets": [
{
"id": "llm-cloud",
"kind": "llm",
"name": "Cloud model",
"description": "Fast general model for non-sensitive requests",
"cost": {
"usd_per_1k_tokens": 0.003
},
"latency_ms": 700,
"quality_prior": 0.8
},
{
"id": "llm-onprem",
"kind": "llm",
"name": "On-prem model",
"description": "Regulated and personal data that must not leave the network",
"cost": {
"usd_per_1k_tokens": 0.0008
},
"latency_ms": 1200,
"quality_prior": 0.7,
"constraints": {
"data_boundary": "on_prem",
"pii_allowed": true
}
}
],
"rules": [
{
"name": "pii-stays-home",
"when": {
"contains_pii": true
},
"prefer": [
"llm-onprem"
],
"pin": true
}
],
"objective": {
"quality": 1,
"cost": 0.3,
"latency": 0.2
}
}
Installed into a catalogue, chosen by a router
Install Privacy-first routing and it becomes one more candidate the router can pick - when it fits.
A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.