One API.
Every model.
One OpenAI-compatible gateway. Cloud providers and local inference. Automatic routing and failover across everything you connect. One API key. One balance. No unnecessary complexity.
Models
Every model served by this gateway, pulled live from the backend. Local models run on your own hardware at $0.
Prepaid credits. Per-token billing.
Credits are consumed at retail rate per token and checked before every request — if you run out, you get a 402 and a top-up link. No subscription, no surprise bills.
Get your API key
One key for every model. Create it, store it safely, top it up when you want cloud inference.
Inference console
Real requests against the live gateway. Pick any model from the catalog and run it. Results come straight from the provider — nothing here is simulated.
Request
Unset controls use the model/Ollama defaults.
Run local models free (bring your own Ollama)
- Install Ollama on your machine (see ollama.com).
- Pull a model, e.g.
ollama pull llama3.1:8b. - Point Clarity at your Ollama host (set the local base URL in the gateway config).
- Pick a Local model above and run it. Billed at $0 — no cloud key needed.
Local inference runs on your hardware, not on Clarity's servers.
Response
OpenAI compatible
Swap your base URL and key. That's it — works with the OpenAI SDK, LangChain, curl, anything that speaks the protocol.
# base URL is shown below — your live gateway import openai client = openai.OpenAI( base_url=…, api_key="gw_…", ) r = client.chat.completions.create( model="gpt-oss-120b", messages=[{"role": "user", "content": "hi"}], ) print(r.choices[0].message.content)
# curl example — replace the key and model curl … \ -H "Authorization: Bearer $KEY" \ -H "Content-Type: application/json" \ -d '{"model":"gpt-oss-120b","messages":[{"role":"user","content":"hi"}]}'
Bring your own local models
Local models run on your machine via Ollama (or any OpenAI-compatible local server) — not on Clarity's cloud. They are billed at $0 through the same API key, with no cloud credentials required.
ollama pull llama3.1:8b. Any model your Ollama supports works.Because inference happens on your hardware, speed and availability depend on your machine — Clarity only routes and meters the request.
Local Models
Checking local Ollama…
Local Runtime
Checking local Ollama runtime…
Round-trip is measured by Clarity around the local upstream request; it is not Ollama model-eval time.
Gateway health
Derived live from the backend — nothing here is hard-coded.
Endpoints
| Endpoint | Auth | Purpose |
|---|---|---|
POST /v1/chat/completions | customer key | OpenAI chat completions (streaming supported) |
GET /v1/models | public | list available canonical models |
GET /v1/usage | customer key | tokens, cost, remaining balance |
POST /v1/signup | public | create a new customer key |
POST /v1/checkout | customer key | Stripe checkout URL to top up credits |
GET /health | — | liveness |
Live gateway status
Pulled live from GET /v1/status — reachable and probe_latency_ms come from a real, zero-cost network probe (any HTTP/TLS response counts, including 401/403). Nothing is hard-coded, estimated, or invented. Latency reflects the probe only; it is not a model inference time.
| Provider | Configured | Creds set | Reachable | Latency (ms) | Models |
|---|
System diagnostics
Read-only, zero-inference. Aggregated from local discovery and the local runtime — nothing here is hard-coded.
— ——Frozen checkpoint is the recovery baseline; the current build may contain newer validated changes. This panel never runs inference and does not claim production readiness.