Build

Routing

Forge can pick the best upstream channel for a request based on price, latency, region, or a fallback chain you define.

Modes

cheapest — minimum cost among eligible channels.
fastest — lowest p50 TTFT in the last 5 minutes.
balanced (default) — weighted score across cost, latency, and recent error rate.
pinned — always use the exact channel set on the key.

Per-request override

bashbash

curl https://api.cytokenhub.com/v1/chat/completions \
  -H "Authorization: Bearer $FORGE_KEY" \
  -H "X-Forge-Route: cheapest" \
  -d '{ "model": "anthropic/claude-4.5-sonnet", "messages": [...] }'

Fallback chains

Provide a comma-separated list of model IDs. Forge tries each in order and falls through on 429 / 5xx / timeout.

httphttp

X-Forge-Route: openai/gpt-5,anthropic/claude-4.5-sonnet,google/gemini-3-pro

Routing rules in the console

Open Channels → Routes in the console to author rules declaratively — match by model glob, region, or header, then choose a channel and a fallback chain. Rules are evaluated top-down.