Quickstart

One call to get a routing decision. No auth required in v1 — the demo router is public.

bash 
curl -X POST https://slng.polsia.app/api/v1/route \
  -H "Content-Type: application/json" \
  -d '{"use_case":"transcription","region":"us","optimize_for":"latency"}'

Response:

json 
{
  "request_id": "3f8a2b1c-4d5e-6f7a-8b9c-0d1e2f3a4b5c",
  "provider": "Deepgram",
  "model": "Nova-3",
  "region_endpoint": "us-east-1",
  "estimated_latency_ms": 180,
  "estimated_cost_per_minute_usd": 0.0043,
  "fallback": ["AssemblyAI Universal-2", "Google STT us-central"],
  "routing": {
    "use_case": "transcription",
    "region": "us",
    "optimize_for": "latency",
    "input_format": null,
    "reason": "Lowest p95 latency in us-east (avg 180ms). Streaming-first, phone-grade noise handling."
  }
}

POST /api/v1/route

Returns the provider slng would route to for the given workload parameters. Each response is logged (provider, params, request_id) for usage analytics.

POST /api/v1/route

Request body

Parameter Type Description
use_case string required
Type of voice workload.
transcription · tts · conversational
region string required
Geographic region for data residency and latency routing.
us · eu · apac
optimize_for string required
Primary optimization target.
latency · cost
input_format string optional
Audio input format. Affects provider selection when transcribing.
wav · mp3 · pcm16

Example — transcription, EU, optimize latency

bash 
curl -X POST https://slng.polsia.app/api/v1/route \
  -H "Content-Type: application/json" \
  -d '{
    "use_case": "transcription",
    "region": "eu",
    "optimize_for": "latency",
    "input_format": "wav"
  }'

Example — TTS, US, optimize cost

bash 
curl -X POST https://slng.polsia.app/api/v1/route \
  -H "Content-Type: application/json" \
  -d '{"use_case":"tts","region":"us","optimize_for":"cost"}'

Example — conversational agent, APAC

bash 
curl -X POST https://slng.polsia.app/api/v1/route \
  -H "Content-Type: application/json" \
  -d '{"use_case":"conversational","region":"apac","optimize_for":"latency"}'

GET /api/v1/providers

Returns the full provider catalog used by the router — names, capabilities, regions, latency benchmarks, and pricing. Useful for building your own tooling on top of the routing data.

GET /api/v1/providers
bash 
curl https://slng.polsia.app/api/v1/providers

Each provider entry:

json 
{
  "id": "deepgram",
  "name": "Deepgram",
  "capability": "stt",
  "regions": ["us-east-1", "eu-west-1", "ap-east-1"],
  "latency_p50_ms": 170,
  "cost_per_minute_usd": 0.0043,
  "notes": "Nova-3 pricing. Source: https://deepgram.com/pricing (May 2026)"
}

Response shape

All responses are JSON. Success responses from /api/v1/route:

request_id string (UUID) Unique ID for this routing decision. Logged to database for usage analytics.
provider string Provider slng would route to (e.g. "Deepgram", "ElevenLabs", "Groq").
model string Specific model at that provider (e.g. "Nova-3", "Turbo v2.5").
region_endpoint string Regional endpoint for data residency (e.g. "us-east-1", "eu-west-1", "ap-east-1").
estimated_latency_ms number Estimated p95 latency in milliseconds based on historical benchmarks for this provider + region.
estimated_cost_per_minute_usd number Estimated cost per audio minute in USD based on provider public pricing.
fallback string[] Ordered list of fallback providers slng would try if the primary fails or rate-limits.
routing.use_case string Echo of request use_case.
routing.region string Echo of request region.
routing.optimize_for string Echo of request optimize_for.
routing.reason string Human-readable explanation of why this provider was selected.

Error responses

400 Bad Request — invalid or missing required parameter. Body: { "error": "..." }

429 Too Many Requests — rate limit exceeded. Body: { "error": "...", "retry_after_seconds": N }. Header: Retry-After: N

Rate limits

30 requests per minute per IP. Tracked in-memory, resets on server restart.

Rate limit headers are included on every response:

X-RateLimit-Limit: 30
X-RateLimit-Remaining: 29

When the limit is exceeded, the response is 429 with a Retry-After header indicating seconds until the window resets.

v1 is a demo router for evaluation — rate limits are intentionally generous. Production access (higher limits, SLA, dedicated routing) requires an API key. Request access →

How routing decisions are made

Every call to POST /api/v1/route runs through the same decision engine slng uses internally. The router evaluates three dimensions:

1

Use case

STT, TTS, and LLM providers have different latency/cost profiles. The router only considers providers in the correct capability bucket — Deepgram is never a TTS fallback.

2

Region

Data residency is a hard constraint. EU traffic routes exclusively to EU endpoints — providers without EU infrastructure are excluded, regardless of latency or cost. Same for APAC. Region selection determines which providers are in scope before any optimization is applied.

3

Optimization target

Within the eligible provider set for a use case + region, the router selects the best option for your stated priority. Latency picks the provider with the lowest p95 response time. Cost picks the cheapest option that still meets quality requirements for the workload.

4

Failover chain

The fallback array is an ordered list of alternatives slng activates if the primary provider rate-limits, returns errors, or exceeds latency thresholds. Failover is automatic — no code change, no incident.

The routing table used by this API is exactly the same one powering the playground. A /compare endpoint (coming soon) will let you see how routing decisions differ across optimization targets and regions side-by-side.