Every model. One way in.
The Claude, GPT and Gemini you already use are here, and you can generate images, video and audio too.
Why teams choose us
Settled on actual usage
Hold → settle on actual usage → full refund when a failure produced no usageEach request reserves funds up front and settles on actual usage, and running out of credits returns 402 instead of overdrawing. Failed requests are not charged; but if the model developer has already charged for what you sent (for example, partial output was produced, a stream was cut off, or a timeout was estimated), the request is charged as usual.
Every call is traceable
request ID → time · model · tokens · spend · speedEvery call gets a request ID. The console shows time, model, tokens, spend and speed; quote the request ID when you contact support.
Managed as a team
Shared wallet → member limits → per-key limitsOne wallet for the whole team. Admins add team members and set per-member limits; each API key can have its own limit, expiry, models and IP rules.
Connect in three minutes
Replace sk-... with an API key from the console.
Coding
Point Claude Code, Codex CLI, Cursor or Cline here and share one bill across the team.
Chat and reasoning
OpenAI and Anthropic formats are supported; existing SDKs only need a new base URL.
Add this to env in ~/.claude/settings.json (or set the same environment variables), then restart Claude Code.
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.relayvia.dev",
"ANTHROPIC_AUTH_TOKEN": "sk-...",
"ANTHROPIC_MODEL": "claude-opus-5"
}
}Three steps
- 1
Open an account
Contact us to open an account, then sign in to the console as admin.
- 2
Create an API key
Create a key in the console with an optional limit, expiry and model list. You can view it again in the list at any time.
- 3
Change the base URL
Point your tool or SDK at us and keep using official model names.
Leave the AI calls to us
Official model names
Use the official model name as-is; no aliases to learn and no config changes when you switch tools.
Prepaid in USD, pay per use
Priced in USD from prepaid credits. Every charge is written to a ledger you can reconcile line by line.
OpenAI and Anthropic formats
OpenAI and Anthropic request formats both work, streaming included.
Control down to each key
Spending limits with reset periods, expiry, model and IP allowlists, and one-click disable for leavers.
Keys stored as hashes
We keep only a hash and the last 4 characters of each key, so a database leak exposes no usable keys.
Human support
Talk to a person. A request ID is enough for us to find the exact call.