Overview

Base URL#

https://api.ahurasense.com/v1

Every request needs an API key in the Authorization header. Keys are created in the dashboard under Inference → API keys and start with ahu_live_. See Authentication.

Endpoints#

Quickstart#

1. Create an account, then open Inference → API keys and create a key. The full key is shown once.
2. Put it in an environment variable and make the request below.

curl https://api.ahurasense.com/v1/chat/completions \
  -H "Authorization: Bearer $AHURA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-haiku-4.5",
    "messages": [
      {"role": "user", "content": "In one sentence, what is a GPU?"}
    ]
  }'

The response is a standard chat completion object:

{
  "id": "chatcmpl-…",
  "object": "chat.completion",
  "model": "claude-haiku-4-5",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "A GPU is a processor built to run thousands of small calculations in parallel, which makes it fast at graphics and at training and running AI models."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 16, "completion_tokens": 34, "total_tokens": 50 }
}

Available models#

The public catalog, live. Prices are input / output per million tokens; models marked hosted run on AhuraSense GPU infrastructure. Context windows, capabilities and cached-input rates are on the Models page, and GET /v1/models returns the same list to your code.

What happens to a request#

Every call passes through the same steps, in this order, at the edge closest to you:

1. Authentication. The key is hashed and looked up. Expired keys, keys called from an address outside their IP allowlist, and unknown keys get 401.
2. Spend cap. If the key or the organization has a monthly hard cap and it is reached, the request stops with 402.
3. Rate limit. A per-key token bucket. Over the limit is 429 with Retry-After.
4. Validation and routing. The body is checked, the model is resolved (directly or through a preset), and the key’s model allowlist is applied.
5. Guardrail. Prompt text is scanned for injection patterns; by default the result is only annotated on the response. Guardrails.
6. Cache. Deterministic, non-streaming requests can be answered from the exact-match cache. Caching.
7. The model. The request is forwarded to the model’s serving backend and the answer is streamed or returned to you unchanged.

Usage is recorded after the response completes, so metering never adds latency to the call itself. Every response carries X-Ahura-Request-Id; quote it when you contact support and we can find the exact request.

Managed in the dashboard#

Some features are configured in the dashboard rather than through the API, and then apply to your requests automatically:

• API keys: model allowlists, IP allowlists, expiry, rate limits, spend caps, zero data retention.
• Presets: named default models you switch without a deploy.
• Usage and audit: per-key and per-model spend, latency percentiles, a CSV export, and an audit trail of every change to your account.
• Batches, fine-tuning, vector collections, notifications: available from the dashboard today; API access to these is on the roadmap.

Where next

Read Chat completions for every parameter, or SDKs & frameworks if you already have code that talks to OpenAI or Anthropic.