Relay API Reference

Base URL

https://api.vibgrate.com/relay/v1

Relay's endpoint shape is OpenAI-compatible (/relay/v1/..., not Vibgrate's usual /v1/<resource> API convention) specifically so existing OpenAI-client tooling can point at it with only a base-URL and key change.

Authentication

Authorization: Bearer <dsn-key>:<dsn-secret>

The credential is the same DSN key/secret pair vg login already provisions — there is no separate Relay-only credential to mint or rotate.

GET /relay/v1/models

Returns the current curated, priced model catalog. No upstream provider call is made to serve this — it works even if a specific upstream provider is degraded. Exactly one entry is flagged default.

curl https://api.vibgrate.com/relay/v1/models \
  -H "Authorization: Bearer $VIBGRATE_RELAY_TOKEN"

POST /relay/v1/chat/completions

OpenAI chat-completions-shaped request and response.

curl https://api.vibgrate.com/relay/v1/chat/completions \
  -H "Authorization: Bearer $VIBGRATE_RELAY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5",
    "messages": [{"role": "user", "content": "..."}],
    "stream": true
  }'

• model — optional; falls back to the catalog default when omitted. See Model selection & routing — this is a one-time choice for the whole session, never a per-request auto-switch.
• stream — supported; usage accounting is read from the stream's final usage chunk once it completes, so streaming never blocks on billing.
• max_tokens — used to compute the conservative cost estimate reserved before the request is served (see Pricing & credits); when omitted, Relay assumes a bounded fallback so an unbounded request can still be estimated.

Error responses

See Troubleshooting & FAQ for how these surface inside VG Code and the panel.