// openai-compatible api proxy

See exactly what
your AI costs.

One API endpoint. Every request logged. Per-model cost tracking down to the token. Drop in your base URL, keep your existing code, and finally know where the money goes.

curl
# Use it like OpenAI. Works with any SDK.
curl https://infersave.polsia.app/v1/chat/completions \
  -H "Authorization: Bearer isk_your_key" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","messages":[{"role":"user","content":"Hello"}]}'
 
# Response includes cost headers
X-InferSave-Cost: 0.000350
X-InferSave-Latency: 1247
What you get

Every token. Every cent. Every request.

$

Per-request cost

Every API call logged with exact token counts and USD cost. No estimates, no averages.

Multi-provider

Route to OpenAI or Anthropic through one endpoint. Same OpenAI-compatible format.

</>

Drop-in replacement

Change your base URL. That's it. Works with the OpenAI SDK, LangChain, or raw HTTP.

Supported models

Pay provider rates. We track for free.

Model Provider Input / 1M tokens Output / 1M tokens
gpt-4o OpenAI $2.50 $10.00
gpt-4o-mini OpenAI $0.15 $0.60
claude-sonnet Anthropic $3.00 $15.00
claude-haiku Anthropic $0.25 $1.25
Quick start

Three steps. Two minutes.

1

Create an API key

POST to /api/keys with a name. You'll get an isk_ key back once.

curl -X POST https://infersave.polsia.app/api/keys \
  -H "Content-Type: application/json" \
  -d '{"name":"my-app"}'
2

Point your SDK at InferSave

Change the base URL. Keep everything else the same.

// Python
client = OpenAI(
  base_url="https://infersave.polsia.app/v1",
  api_key="isk_your_key_here"
)

// Node.js
const client = new OpenAI({
  baseURL: "https://infersave.polsia.app/v1",
  apiKey: "isk_your_key_here"
})
3

Make requests as usual

Use gpt-4o, gpt-4o-mini, claude-sonnet, or claude-haiku. Anthropic models are automatically translated to the Messages API format.

Know your AI costs.
Down to the token.

No dashboards to configure. No SDKs to install. Just change your base URL and every request is tracked.