Frequently asked questions
Is it really OpenAI-compatible?
Yes. Point any OpenAI SDK at https://api.aix.theaimart.co/v1 with your AIx key and existing
code works unchanged. See Migrating from OpenAI — it’s a
two-line change (base URL + key).
Which models can I use?
Whatever is in the live catalog. Read it from GET /v1/models or browse the models page — chat, embeddings, image, TTS and transcription models behind one AIx contract. The catalog changes without an API version bump, so don’t hardcode ids.
How does billing work?
AIx is prepaid: 1 credit = $1. Each call holds a token-based estimate, then settles on the route’s reported usage and refunds the difference. You only pay for what’s actually used. Full detail in Billing model. Top up with Razorpay or PayPal from billing.
What are the rate limits?
Each key is limited to 100 requests per minute. You can additionally set your own
daily/monthly spend caps and tokens-per-minute (TPM) limits per key — see
Limits, caps & scopes. Hitting a limit returns 429 rate_limit_error;
back off and retry.
Do you store my prompts?
No. We record only model, token/unit counts, cost and a timestamp per call — never your prompts or the model’s responses. Keys are stored hashed. See Data & privacy.
Your prompt is sent to the upstream provider that serves each call, for the duration of that request — their policy applies to that hop. AIx adds no retention of its own.
Can I stream responses?
Yes — set stream: true for text/event-stream chunks, terminated by data: [DONE] on the
chat-completions surface. See Streaming. The Messages surface emits its
documented event lifecycle instead.
Do tools / JSON mode / vision work?
Yes, on eligible models. AIx restricts tools, tool_choice, response_format and
multimodal requests to execution paths that support the complete payload. See the
compatibility layer.
What happens if an execution route goes down?
The request transparently fails over to the next eligible route, and a circuit breaker stops
sending traffic to a failing path for a cooldown. You only see an error
(503 service_unavailable) if every candidate fails. See Routing & failover.
Is there an SLA / status page?
See the live status page and Support & SLA.
How do I avoid double-charging on retries?
For POST /v1/chat/completions, send an Idempotency-Key header. A repeated non-streaming
call with the same key (within 24h) replays the completed result.