Build the thing.We’ll handle the models.
The model market moves weekly. Your integration shouldn’t.
No subscription. Ship today, pay only for what you use.
Shipping this week
Input / 1M tokens| Model | Provider | Context | Input / 1M | Typical p50 |
|---|---|---|---|---|
| Claude Fable 5 | Anthropic | 1M | $10.00 | 220ms |
| GPT-5.6 | OpenAI | 272K | $5.00 | 220ms |
| Grok 4.5 | xAI | 500K | $2.00 | 210ms |
| DeepSeek V4 | DeepSeek | 1M | $0.43 | 190ms |
| o3 | OpenAI | 200K | $10.00 | 850ms |
| Gemini 2.5 Pro | 1M | $1.25 | 310ms |
How it works
You ship the feature. We run the request.
One OpenAI-compatible call. Cap, route, failover, and a receipt—handled below your code, so model churn never becomes a rewrite.
You set the boundary once
Scoped to a project, with a model allow-list, rate limit, and a spend ceiling that hard-stops at the cap—so a bad loop can’t empty the wallet.
We pick the provider
Your rules choose the path. If a provider fails, the request retries the next candidate in under 200ms—without you rewriting a client.
The answer explains itself
Every response carries cost and routing: which provider served it, which rule matched, and why. No black box between you and production.
Spend stays attributable
Each request lands in your usage log by key, agent, and model—so “where did the money go?” has an answer, not a guess.
Platform
Ship like the models are already sorted
Retries, failover, and spend limits live under the API. Production traffic shouldn't force you to re-architect for the next model drop.
When a provider blinks, you don't
Relixr retries the next-best option and records the chain—the rule that matched, the candidates considered, and why the winner won. The routing record is part of the response, so you can ship without babysitting the upstream.
Keep your SDK
Point any OpenAI-compatible client at Relixr. LangChain, LlamaIndex, or a raw fetch—same shape you already ship.
Swap models, not code
Change the model id (or a routing rule) when something better lands. Your integration stays put.
Hard spend caps
Balance is reserved before each request. Hit the ceiling and the call is refused—never billed past it.
Failover without drama
Provider 5xx? Next candidate retries in under 200ms, with the chain recorded on the response.
Cost you can explain
Every request is tagged to a key, agent, and model—so finance questions don’t wait on a spreadsheet.
Hit the limit mid-PR?
Keep Cursor, Claude Code, or Codex. Point them at Relixr, top up any amount, and finish the change with a hard spend cap the subscription never gave you.
Billing
Spend only on what ships
Start free, then add prepaid credits when you need frontier models—no subscription standing between you and the model you need this week.
Balance is reserved before each request, so you never charge past what you hold. Credits don't expire while you iterate.
- Getting started
- FreeAccount + community models
- Minimum balance
- $0.00No minimum top-up
- Credit expiry
- NeverUse at your own pace
- Payment
- Card or localLocal options for United States