AI

The AI API provides developers with seamless access to AI functionality without requiring API keys, configuration, or extra dependencies.

Some users might not have access to this API. If a user doesn't have access to Raycast Pro, they will be asked if they want to get access when your extension calls the AI API. If the user doesn't wish to get access, the API call will throw an error.

You can check if a user has access to the API using environment.canAccess(AI).

API Reference

AI.ask

Ask AI anything you want. Use this in “no-view” Commands, effects, or callbacks. In a React component, you might want to use the useAI util hook instead.

Signature

async function ask(prompt: string, options?: AskOptions): Promise<string> & EventEmitter;

Example

Basic Usage

import { AI, Clipboard } from "@raycast/api";

export default async function command() {
  const answer = await AI.ask("Suggest 5 jazz songs");

  await Clipboard.copy(answer);
}

Error handling

import { AI, showToast, Toast } from "@raycast/api";

export default async function command() {
  try {
    await AI.ask("Suggest 5 jazz songs");
  } catch (error) {
    // Handle error here, eg: by showing a Toast
    await showToast({
      style: Toast.Style.Failure,
      title: "Failed to generate answer",
    });
  }
}

Stream answer

import { AI, getSelectedFinderItems, showHUD } from "@raycast/api";
import fs from "fs";

export default async function main() {
  let allData = "";
  const [file] = await getSelectedFinderItems();

  const answer = AI.ask("Suggest 5 jazz songs");

  // Listen to "data" event to stream the answer
  answer.on("data", async (data) => {
    allData += data;
    await fs.promises.writeFile(`${file.path}`, allData.trim(), "utf-8");
  });

  await answer;

  await showHUD("Done!");
}

User Feedback

import { AI, getSelectedFinderItems, showHUD } from "@raycast/api";
import fs from "fs";

export default async function main() {
  let allData = "";
  const [file] = await getSelectedFinderItems();

  // If you're doing something that happens in the background
  // Consider showing a HUD or a Toast as the first step
  // To give users feedback about what's happening
  await showHUD("Generating answer...");

  const answer = await AI.ask("Suggest 5 jazz songs");

  await fs.promises.writeFile(`${file.path}`, allData.trim(), "utf-8");

  // Then, when everythig is done, notify the user again
  await showHUD("Done!");
}

Check for access

import { AI, getSelectedFinderItems, showHUD, environment } from "@raycast/api";
import fs from "fs";

export default async function main() {
  if (environment.canAccess(AI)) {
    const answer = await AI.ask("Suggest 5 jazz songs");
    await Clipboard.copy(answer);
  } else {
    await showHUD("You don't have access :(");
  }
}

Parameters

Name	Description	Type
prompt*	The prompt to ask the AI.	string
options	Options to control which and how the AI model should behave.	AI.AskOptions

Return

A Promise that resolves with a prompt completion.

Types

AI.Creativity

Concrete tasks, such as fixing grammar, require less creativity while open-ended questions, such as generating ideas, require more.

type Creativity = "none" | "low" | "medium" | "high" | "maximum" | number;

If a number is passed, it needs to be in the range 0-2. For larger values, 2 will be used. For lower values, 0 will be used.

AI.Model

The AI model to use to answer to the prompt. Defaults to AI.Model["OpenAI_GPT3.5-turbo"].

Enumeration members

Model	Description
OpenAI_GPT4	GPT-4 is OpenAI's most capable model with broad general knowledge, allowing it to follow complex instructions and solve difficult problems.
OpenAI_GPT4-turbo	GPT-4 Turbo from OpenAI has a big context window that fits hundreds of pages of text, making it a great choice for workloads that involve longer prompts.
OpenAI_GPT4o	GPT-4o is the most advanced and fastest model from OpenAI, making it a great choice for complex everyday problems and deeper conversations.
OpenAI_GPT4o-mini	GPT-4o mini is a highly intelligent and fast model that is ideal for a variety of everyday tasks.
Anthropic_Claude_Haiku	Claude 3 Haiku is Anthropic's fastest model, with a large context window that makes it ideal for analyzing code, documents, or large amounts of text.
Anthropic_Claude_Sonnet	Claude 3.5 Sonnet from Anthropic has enhanced intelligence with increased speed. It excels at complex tasks like visual reasoning or workflow orchestrations.
Anthropic_Claude_Opus	Claude 3 Opus is Anthropic's most intelligent model, with best-in-market performance on highly complex tasks. It stands out for remarkable fluency.
Perplexity_Llama3.1_Sonar_Small	Perplexity's Llama 3.1 Sonar Small is built for speed. It quickly gives you helpful answers using the latest internet knowledge while minimizing hallucinations.
Perplexity_Llama3.1_Sonar_Large	Perplexity's advanced model. Can handle complex questions. It considers current web knowledge to provide well-reasoned, in-depth answers.
Perplexity_Llama3.1_Sonar_Huge	Perplexity's most advanced model. Offers performance that is on par with state of the art models today.
Llama3.1_70B	Llama 3.1 70B is a versatile open-source model from Meta suitable for complex reasoning tasks, multilingual interactions, and extensive text analysis. Powered by Groq.
Llama3.1_8B	Llama 3.1 8B is an open-source model from Meta, optimized for instruction following and high-speed performance. Powered by Groq.
Llama3_70B	Llama 3 70B from Meta is a highly capable open-source LLM that can serve as a tool for various text-related tasks. Powered by Groq.
Llama3.1_405B	Llama 3.1 405B is Meta's flagship open-source model, offering unparalleled capabilities in general knowledge, steerability, math, tool use, and multilingual translation. Powered by together.ai
MixtraL_8x7B	Mixtral 8x7B from Mistral is an open-source model that demonstrates high performance in generating code and text at an impressive speed. Powered by Groq.
Mistral_Nemo	Mistral Nemo is a small model built in collaboration with NVIDIA, and released under the Apache 2.0 license.
Mistral_Large2	Mistral Large is Mistral's flagship model, capable of code generation, mathematics, and reasoning, with stronger multilingual support.

If a model isn't available to the user, Raycast will fallback to a similar one:

• AI.Model.Anthropic_Claude_Opus and AI.Model.Anthropic_Claude_Sonnet -> AI.Model.Anthropic_Claude_Haiku

• AI.Model.OpenAI_GPT4 and AI.Model["OpenAI_GPT4-turbo"] -> AI.Model["OpenAI_GPT4o-mini"]

• AI.Model["Perplexity_Llama3.1_Sonar_Large"] and AI.Model["Perplexity_Llama3.1_Sonar_Huge"] -> AI.Model["Perplexity_Llama3.1_Sonar_Small"]

• AI.Model.Mistral_Large2 -> AI.Model.Mistral_Nemo

AI.AskOptions

Properties

Property	Description	Type
creativity	Concrete tasks, such as fixing grammar, require less creativity while open-ended questions, such as generating ideas, require more. If a number is passed, it needs to be in the range 0-2. For larger values, 2 will be used. For lower values, 0 will be used.	AI.Creativity
model	The AI model to use to answer to the prompt.	AI.Model
signal	Abort signal to cancel the request.	AbortSignal