Skip to main content

How it works

How GonkaGate routes requests to Gonka Network

Routing architecture and health signals that drive node selection in Gonka Network, from client to response.

  • Routing + fallbackone decision
  • Signalsavailability + latency
  • Metadatacost + latency per response

Get a key to try routing live. The Gonka API overview has the full surface and compatibility details.

Detailed API surface lives on the Gonka API page.

Routing snapshot

Routing guarantees at a glance

Key outcomes you can rely on per request.

Primary goal
Fast first byte
Retries
Up to x3 per request
Switch signal
No data in 10 s
Visibility
Usage metadata per response

Who it's for

Teams that need predictability

Built for stable streaming with bounded retries.

  • API teamsStable streaming, bounded retries
  • Product leadersLower perceived latency, higher reliability
  • No-code buildersPredictability without routing tuning

Access check

Verify API key status and account access.

Route selection

Pick the best node for the selected model.

Fallback handling

Re-route on capacity or health issues.

Usage logging

Capture cost, latency, and token metadata.

Response delivery

Return payload with usage fields.

Request flow

Client
Gateway
Router
Network
Response

Routing decisions use availability and latency signals.

Why routing stays fast

Smart routing per model

We score routes per model and pick the one most likely to respond fast. Reliability stays high when the network is unstable.

Pool
Per-model route pool
Scoring
Recent success + latency
Safety
Unstable routes cool off
Result
Fast, reliable route now

Guardrails

Speed with safe recovery

If the first byte is slow, we switch quickly and keep retries bounded.

First-byte timeout

If no data arrives in 10 s, stop and switch routes.

  • Attempt 1no data in 10 s, stop
  • Attempt 2first byte, continue stream
Retry policy

Up to x3 attempts, always with a different route.

  • Retry ontimeout or 429 (rate limit)
  • Isolationeach attempt uses a new route
Route lifecycle

New routes warm up first; repeated failures go to quarantine.

New — Warmup — HealthyHealthy — Fail/slow x3 — Quarantine, 5 m backoffRelease — Healthy

Operational rules

  • Slow first byteweak signal; hard errors are strong signals
  • No repeata route is not reused within one request
  • Fallbackif healthy routes are missing, use cached endpoints excluding quarantine