Call AI models

Send AI requests with the user's access token. rNet forwards them to the model provider and charges the user's credits.

Make a request

Create a ModelClient with a model ID, then call chat with a request body and the access token.

javascript

const response = await ai.chat({
  contents: [{ role: 'user', parts: [{ text: 'Write a haiku about credits.' }] }],
}, accessToken);

console.log(response);

Request bodies

rNet passes your body to the provider unchanged, so send the provider's own request format and you get the provider's response back.

Models	Body format
Gemini and Gemma (Google)	Gemini `generateContent` body with `contents`
`gemini-embedding-001`	Gemini `embedContent` body with `content`
`text-embedding-3-small`	OpenAI embeddings body with `model` and `input`
Llama, Mixtral, Gemma 2 (Groq)	OpenAI-style chat body with `model` and `messages`

javascript

// Groq model: OpenAI-style chat completions body
const groq = new ModelClient('llama-3.1-8b-instant');
const reply = await groq.chat({
  model: 'llama-3.1-8b-instant',
  messages: [{ role: 'user', content: 'Hello!' }],
}, accessToken);

// OpenAI embeddings
const embed = new ModelClient('text-embedding-3-small');
const vectors = await embed.chat({
  model: 'text-embedding-3-small',
  input: 'rNet credits work in any app',
}, accessToken);

Stream responses

Use chatStream (chat_stream in Python) to receive the response as it is generated. For Groq models, also set "stream": true in the body.

javascript

const stream = await ai.chatStream(payload, accessToken); // ReadableStream
const reader = stream.getReader();
const decoder = new TextDecoder();

for (;;) {
  const { value, done } = await reader.read();
  if (done) break;
  res.write(decoder.decode(value)); // forward server-sent events to your client
}
res.end();

Handle errors

The SDKs throw when rNet returns an error. The message contains the HTTP status, for example Request failed: 402 - Insufficient funds.

javascript

try {
  const response = await ai.chat(payload, accessToken);
} catch (err) {
  // SDK errors look like: "Request failed: 402 - Insufficient funds"
  const status = Number(/Request failed: (\d{3})/.exec(err.message)?.[1]);

  if (status === 401) {
    // Access token expired: refresh it and retry once
  } else if (status === 402) {
    // The user is out of credits: ask them to top up at rnetai.org/user-home
  } else if (status === 429) {
    // Too many requests: wait and retry
  } else {
    throw err;
  }
}

Status	What to do
401	Access token expired or invalid: refresh it and retry once.
402	User is out of credits: ask them to top up in their Credit Wallet.
422	Unknown model ID or a model this app can’t use: check the model list.
429	Too many requests: wait and retry.

See Errors for every status code.

Protect your users' data

Run prompts through Context Eraser to remove emails, keys and other private data before they reach a model.