assisters

Official TypeScript SDK for the Assisters API.

One API key, one client, every model. Typed end to end, no runtime dependencies — just fetch.

Install

npm install assisters

Node 18+ (the SDK uses the built-in fetch).

Quick start

import { AssistersClient } from "assisters";

const client = new AssistersClient(process.env.ASSISTERS_API_KEY!);

const completion = await client.chat.create({
  model: "assisters-chat-v1",
  messages: [{ role: "user", content: "Explain prompt caching in one sentence." }],
});

console.log(completion.choices[0].message.content);

Get a key from the dashboard.

Resources

The client exposes one namespace per API area:

| Namespace | Method | Does | |---|---|---| | client.chat | create(params) | Chat completions | | client.embeddings | create(params) | Vectors for search, clustering, RAG | | client.moderations | create(params) | Content-safety classification | | client.rerank | create(params) | Order documents by relevance | | client.translate | create(params) | Translate text | | client.models | list(), retrieve(id) | The catalogue, with pricing | | client.images | generate(params) | Text to image | | client.audio | speech(), transcribe(), enhance() | TTS, STT, cleanup | | client.webSearch | query(params) | Live web search with citations | | client.usage | tokens(), cost(), rateLimit() | What you have spent |

const { data } = await client.embeddings.create({
  model: "assisters-embed-v1",
  input: "The quick brown fox",
});

const results = await client.webSearch.query({ query: "WCAG 2.2 contrast rules" });

const audio = await client.audio.speech({ input: "Hello there", voice: "nova" });
await writeFile("out.mp3", Buffer.from(audio));

Not here, on purpose

credits, collections, notifications and apiKeys used to be namespaces on this client. Every one of those endpoints authenticates with a signed-in browser session rather than a Bearer key, so they returned 401 for any client built the way this one is. Wallet, collections, notifications and key management live in the dashboard. A marketplace search namespace was also removed — it is served by assisters.io, not by api.assisters.dev.

Errors

Every failure throws a typed error, so you can branch on the cause rather than parse a message:

import { AssistersClient, RateLimitError, AuthError } from "assisters";

try {
  await client.chat.create({ model: "assisters-chat-v1", messages });
} catch (err) {
  if (err instanceof RateLimitError) {
    // err.retryAfterMs tells you how long to wait
  } else if (err instanceof AuthError) {
    // key is missing, invalid or revoked
  }
}

AssistersError is the base class and carries the HTTP status; RateLimitError, AuthError and NotFoundError extend it.

Configuration

const client = new AssistersClient(apiKey, {
  baseUrl: "https://api.assisters.dev", // override for a private deployment
});

Billing

Usage is billed per token against a shared prepaid wallet — no subscription. Rates are per model and published at assisters.dev/pricing; GET /v1/models returns the same figures the meter charges. Prompt-cache reads are billed at 10% of the input rate and cache writes at 1.25x.

Links

• Documentation
• API reference
• Pricing
• Contact

License

MIT