Prerequisites
- The Vibgrate CLI installed (
vgon yourPATH). - A Vibgrate account (
vg login) — Relay is billed per workspace, not per machine. - Prepaid credit on the workspace, or a BYOK provider key (see BYOK passthrough) if you'd rather not spend Relay credit yet.
1. Sign in
vg login
This opens a browser flow and stores a DSN credential locally. Relay authenticates with the same credential — there is no separate Relay-only login.
2. Add credit (skip this if you're using BYOK)
Open dash.vibgrate.com → Settings → Billing → Relay and buy a top-up ($25 minimum). Credits are prepaid — nothing is served on Vibgrate's key unless the balance can already cover it, so there's never a surprise invoice. See Pricing & credits for the full fee schedule.
3. Set your Relay token
export VIBGRATE_RELAY_TOKEN="<your-dsn-key>:<your-dsn-secret>"
Put this in your shell profile or CI secret store — never commit it. When VIBGRATE_RELAY_TOKEN is set, VG Code prefers Vibgrate Relay automatically over other hosted providers, with local models as fallback if Relay is unreachable.
4. Run VG Code
vg code "extract the pagination logic in UserList.tsx into a hook" --provider vibgrate-relay
The --provider flag is optional once VIBGRATE_RELAY_TOKEN is set — Relay is picked automatically. Pick a specific model with --model or VG_CODE_MODEL; see Supported models for the current catalog.
vg code "..." --provider vibgrate-relay --model anthropic/claude-sonnet-5
Verifying it's working
vg doctor
vg doctor reports which model backend VG Code would currently select and why (the same routing reason string VG Code itself uses, e.g. "VIBGRATE_RELAY_TOKEN is set → Vibgrate Relay (with local fallback)").
What happens if you have no credit and no BYOK key
VG Code fails with an actionable message rather than a silent local fallback substitution of a different paid backend:
no model backend configured. Do one of: sign in to Vibgrate Relay (VIBGRATE_RELAY_TOKEN),
set OPENROUTER_API_KEY for the hosted router, run `vg models install spark` for the
Vibgrate manager, run Ollama locally, or use --local with a local model.
A request against an empty Relay balance returns HTTP 402 with an insufficient_credit error carrying a top-up link and a correlation id — see Troubleshooting & FAQ.