About VaultAPI
VaultAPI is an AI gateway for people who pay for their own tokens. One endpoint in front of every provider, your keys encrypted at rest, and honest numbers on what each request actually costs.
Adding a second LLM provider to an app is never just an API call. It's another SDK, another key in another secret store, another retry policy, and another invoice arriving from another vendor. By the third provider, nobody can answer the simple question: what did this feature cost us last week?
VaultAPI collapses that into one OpenAI-compatible endpoint. You keep your own provider accounts and your own rates — we sit in the middle to route, count, cost and log. The result is that switching models becomes a rule instead of a refactor, and your spend becomes a chart instead of a surprise.
VaultAPI is bring-your-own-key by design. We never resell inference or mark up tokens. Your provider keys are envelope-encrypted with AES-GCM the moment they arrive, and decrypted only for the instant a request needs them — never written to a log, a response, or a disk.
Most teams discover their AI spend at the end of the month. We think you should see it before you send the request: provider-exact token counts, a cost estimate across every model, and per-request logs afterwards so nothing is a mystery.
Most prompts don't need your most capable model. Routing rules send the easy majority to cheaper models automatically, and the advisor mines your real traffic to suggest new rules with the savings it projects.
We speak the OpenAI-compatible API you already use. Adopting VaultAPI is a base_url change; leaving is the same change in reverse. Your data, keys and models remain yours throughout.
Encryption
AES-GCM envelope
Per-key data keys wrapped by a versioned master key, rotatable without re-encrypting payloads.
Isolation
Per-workspace
Every key, rule, budget and log is scoped to your organization and checked on every request.
Portability
OpenAI-compatible
Standard request and response shapes, so your existing SDKs and tooling keep working.
No card up front. Add a provider key, send one request, and watch the first log line appear.