Use your developer API key as a bearer token. Optional HMAC signatures can be enforced with OMIXA_REQUIRE_SIGNATURE=true.

Authorization: Bearer omx_live_xxx
Content-Type: application/json
X-Omixa-Timestamp: 1780660800
X-Omixa-Signature: sha256=<hmac_sha256(timestamp.method./api/v1/path.raw_body)>

Method	Path	Purpose
GET	/api/v1/models	List enabled models with modality, provider, and prices.
GET	/catalog	Visual developer catalog inside the dashboard.

curl -X POST https://www.omixa.cloud/api/v1/chat/completions \
  -H "Authorization: Bearer omx_live_xxx" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"Hello"}]}'

curl -N -X POST https://www.omixa.cloud/api/v1/chat/completions \
  -H "Authorization: Bearer omx_live_xxx" \
  -H "Accept: text/event-stream" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o-mini","stream":true,"messages":[{"role":"user","content":"Write a short test."}]}'

When stream is true, Omixa returns OpenAI-compatible SSE frames as they arrive. Azure frames are passed through directly, while Vertex Gemini uses native streamGenerateContent and is normalized into chat.completion.chunk events with the final data: [DONE] marker.

For the lowest first-token latency, keep AZURE_STREAM_CHUNK_BYTES=128, AZURE_STREAM_HEARTBEAT_SECONDS=1, and AZURE_STREAM_FLUSH_PADDING_BYTES=8192. Omixa sends SSE comment heartbeats while providers are still thinking, so proxies and browsers do not treat the connection as idle.

In production, disable response buffering, gzip, and HTTP/3/QUIC for /api/v1/chat/completions and /playground/chat/completions at the CDN/proxy layer. The Laravel response also sets X-Accel-Buffering: no, Cache-Control: no-transform, and Alt-Svc: clear.

Reasoning speed is normalized per Azure model family before the request reaches Azure. Original gpt-5 models support minimal, gpt-5.1 supports none, and newer models such as gpt-5.4 and gpt-5.5 support none through xhigh. gpt-5-pro only supports high.

{
  "model": "gpt-5.5",
  "stream": true,
  "reasoning_effort": "xhigh",
  "verbosity": "medium",
  "service_tier": "priority"
}

service_tier: "priority" is Azure Priority processing. Enable Priority processing on eligible Global Standard or Data Zone Standard deployments in Microsoft Foundry, or send service_tier per request. If the deployment does not support Priority processing, turn off Fast mode and Omixa will omit the priority request.

curl -X POST https://www.omixa.cloud/api/v1/embeddings \
  -H "Authorization: Bearer omx_live_xxx" \
  -H "Content-Type: application/json" \
  -d '{"model":"text-embedding-3-small","input":"Knowledge base text"}'

curl -X POST https://www.omixa.cloud/api/v1/images/generations \
  -H "Authorization: Bearer omx_live_xxx" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-image-1","prompt":"A black and white futuristic API dashboard","n":1}'

Method	Path	Purpose
GET	/api/v1/wallet	Current balance and locked balance.
POST	/api/v1/wallet/top-up	Manual/demo top-up endpoint.
GET	/api/v1/wallet/transactions	Full wallet movement history.

Method	Path	Purpose
GET	/api/v1/usage/requests	Per-request logs, tokens, cost, and latency.
GET	/api/v1/usage/rollups	Aggregated hourly and daily usage.
GET	/usage	Visual dashboard reports and live polling.

{
  "error": {
    "code": "insufficient_balance",
    "message": "Wallet balance is not enough for this request."
  }
}

Add many deployments for the same model to distribute load across Azure resources, regions, and keys. Omixa checks RPM, TPM, and hourly token windows before routing.

php artisan omixa:add-azure-deployment gpt-4o-mini eastus-key-1 \
  https://YOUR_RESOURCE.openai.azure.com gpt-4o-mini AZURE_KEY_HERE \
  --region=eastus --rpm=120 --tpm=200000 --hourly-tokens=1000000