astralaneSave on AI models.
Access the newest OpenAI and Anthropic models through one OpenAI-compatible API. Keep your request format. Pay less per token.
- OpenAI-compatible gateway
- Prepaid usage
- No monthly commitment
// One integration. Your routing constraints. const client = new OpenAI({ baseURL: "https://api.astralane.ai/v1", apiKey: process.env.ASTRALANE_API_KEY, }); const response = await client.chat .completions.create({ model: "auto/balanced", messages: [{ role: "user", content: "Hello, world." }], });
Live configuration from the public API · counts, not performance claims
The list price, and what you pay.
Every model is priced against the vendor's published rate. Compare the discount, check current activity, then open a model for its full price history.
No unverified discounts. A savings comparison needs a documented list price. Development models are always labeled and do not represent real vendor pricing.
Start with the supported API subset.
Astralane implements selected OpenAI-compatible chat features and narrower text-only Responses and Messages endpoints. Configuration examples below are not certification of complete SDK, IDE or agent workflows.
import OpenAI from 'openai'; const client = new OpenAI({ apiKey: process.env.ASTRALANE_API_KEY, baseURL: 'https://api.astralane.ai/v1',}); const result = await client.chat.completions.create({ model: 'openai/gpt-5.6-sol', messages: [{ role: 'user', content: 'Hello, world.' }],});console.log(result.choices[0]?.message.content);Configuration example. Verify against your client version. The supplied local integration suite covers OpenAI JavaScript chat streaming and non-streaming with mock supply. Other SDK methods and real model routes need separate verification.
- OpenAI JavaScript: tested chat subset
- Responses: stateless text subset
- Messages: text subset
- Other client configurations: unverified
Not just a gateway.
A better decision layer.
Astralane filters out routes that do not meet your requirements before comparing price, latency, and reliability.
- Constraints first. Optimization second.
Provider, region, budget, tools, and privacy requirements stay in control.
- Failover without an integration rewrite.
Retry eligible failures before output begins. Never replay a partially delivered answer.
- Metadata, not your conversations.
Prompt and response bodies are not stored in the main database.
ROUTING ILLUSTRATION · NOT LIVE TELEMETRY
One wallet. Exact accounting.
Reserve funds before a request. Settle against actual token usage. Inspect the charge and its ledger entries afterward.
Let auto find the route.
Use auto, auto/fast, auto/balanced, or auto/best to select an eligible model using transparent, replaceable routing heuristics.
Privacy is a requirement.
ZDR requests can use only capacity explicitly configured with a matching contract. No qualifying route means no request.
Price it against your own volume.
Pick a model, set the tokens you expect to spend in a month, and compare the vendor list price with the Astralane catalog price.
Estimate only. Cached input is billed at a lower rate and is not counted here. Actual charges follow measured token usage.
Your next request has options.
One workspace. One API key. One place to understand your inference spend.