"no model backend configured"
VG Code couldn't resolve any provider — Relay, another hosted provider, or a local model. Fix any one of:
export VIBGRATE_RELAY_TOKEN="<key>:<secret>" # sign in to Vibgrate Relay
export OPENROUTER_API_KEY="..." # or the hosted OpenRouter router
vg models install spark # or install a local model
Or pass --local with a locally discovered model (Ollama, LM Studio, etc.).
"insufficient_credit" (HTTP 402)
Your workspace's Relay balance can't cover the estimated cost of the request, so nothing was sent upstream and nothing was charged. Either:
- Add credit at
dash.vibgrate.com/settings/relay(minimum $25 top-up), or - Switch to BYOK with your own provider key, which never touches Relay credit — see BYOK passthrough, or
- Fall back to local Code Modes, which cost nothing and need no account.
The error response carries a correlation id — include it if you contact support about a specific request.
Why did my request 503?
The model you requested has no Workers AI mapping and no upstream provider key is configured for it on Relay's side. This fails closed rather than silently losing your request or serving a different model you didn't ask for. GET /relay/v1/models still works during this — only chat/completions for that specific model is affected.
Why did my request 502?
The upstream provider itself returned an error or a non-2xx response. Relay releases your entire reservation before returning the error — you are never charged for a request that produced no tokens. Retry, or try a different model.
Does Relay ever silently switch my model?
No. See Model selection & routing — you choose the model once per session; Relay never substitutes a cheaper or faster model based on cost, latency, or load.
Is my code sent anywhere Relay doesn't tell me about?
Relay proxies your request to whichever backend serves your chosen model (Cloudflare Workers AI directly, or the upstream provider) — never a third destination, and provider keys (yours, on BYOK, or Vibgrate's own) are held in an encrypted vault and resolved per request, never logged.
Can I see exactly what I was charged and why?
Yes — the monthly statement and every billing email are generated from the same ledger and usage rows the dashboard reads, so any charge can be reconciled line-by-line. See Notifications & billing email.
Does Relay work if I don't have a Vibgrate Cloud account?
No — Relay is a hosted, authenticated service billed per workspace, so it requires vg login. If you want a fully offline path with no account, use local Code Modes (--local) instead; Vibgrate AI Context (vg serve) itself remains fully local and account-free regardless of which model backend you use for VG Code.