Mistral AI Edge

A TypeScript module for querying Mistral's API using fetch (a standard Web API) instead of axios.

This is not (at least not currently) a drop in replacement for the official @mistralai/mistralai module (which has axios as a dependency). However, it should be simple enough to switch.

As well as reducing the bundle size, removing the dependency means we can query Mistral from edge environments. Edge functions such as Next.js Edge API Routes are very fast and allow streaming data to the client.

Installation

yarn add mistral-edge

or

npm install mistral-edge

or

bun install mistral-edge

Usage

import { streamMistralChat } from "mistral-edge";

async function main() {
  const tokenStream = streamMistralChat(
    [{ role: "user", content: "Write a haiku about cheese" }],
    {
      model: "mistral-medium",
      temperature: 0.7,
    },
  );

  let fullResponse = "";
  for await (const token of tokenStream) {
    fullResponse += token;
    process.stdout.write(token);
  }

  console.log("\n\nFull output:")
  console.log(fullResponse)
}

main();

Save the above code to a file named example.ts and run ith with the following command:

export MISTRAL_API_KEY=<your API key> && npx ts-node example.ts

Parameters

For details of all parameters, see the official Mistral API docs.

Edge route handler examples

:warning: TODO