AI

window.agentlet.ai gives direct access to an AI provider (OpenAI today) with support for text, images, and PDF documents.

Check availability

const isAvailable = window.agentlet.ai.isAvailable();

Send a text prompt

const response = await window.agentlet.ai.sendPrompt('Analyze this data and provide insights');

sendPrompt resolves to the assistant’s raw text reply. It throws if no provider is configured.

Send a prompt with images

const images = [screenshotBase64, documentImage]; // Data URLs, http(s) URLs, or bare base64 strings
const analysis = await window.agentlet.ai.sendPrompt('What do you see in these images?', images);

Send a prompt with a PDF document

const fileInput = document.getElementById('pdfFile');

const pdfAnalysis = await window.agentlet.ai.sendPromptWithPDF(
    'Analyze this PDF document and summarize the key points',
    fileInput.files[0], // File, ArrayBuffer, Uint8Array, or an http(s) URL string
    {
        pdfOptions: {
            scale: 1.5,   // Higher resolution
            maxPages: 10, // Limit pages for efficiency
            format: 'image/png',
        },
        showInConsole: true, // Display converted images in console (default true)
    }
);

sendPromptWithPDF converts the PDF to images internally, then behaves like sendPrompt.

Convert a PDF to images without sending it to AI

const pdfImages = await window.agentlet.ai.convertPDFToImages(fileInput.files[0]);
// Resolves to an array of base64 data URL image strings, one per page

Status and providers

const status = window.agentlet.ai.getStatus();
console.log(status.currentProvider);              // 'openai'
console.log(status.available);                    // true or false
console.log(status.pdfSupport.available);          // true or false
console.log(status.pdfSupport.capabilities.maxRecommendedPages); // 10

const providers = window.agentlet.ai.getAvailableProviders();
window.agentlet.ai.setProvider('openai');

Validate the current configuration against the provider, useful for a settings screen:

const result = await window.agentlet.ai.validateAPI();
if (result.success) {
    console.log(result.details.model, result.details.responseTime);
} else {
    console.error(result.error);
}

Configuring credentials

AI credentials are read from environment variables managed by window.agentlet.env. See Environment variables for the full API.

window.agentlet.env.OPENAI_API_KEY = 'sk-...';
window.agentlet.env.OPENAI_MODEL = 'gpt-6-luna'; // Optional, this is the default
window.agentlet.ai.refresh(); // Refresh after env changes

OPENAI_MODEL is optional. Since agentlet-core 2.4.0, the default model is gpt-6-luna (earlier releases default to gpt-4o-mini). Set OPENAI_MODEL to use another model, including an older one such as gpt-4o-mini.

The provider adapts the request to the model. Classic models such as gpt-4o-mini receive max_tokens and temperature. The gpt-5, gpt-6 and o-series models receive max_completion_tokens instead, and no temperature (they reject it while reasoning is on). gpt-6-luna and gpt-6-sol are sent reasoning_effort: "none" so the token budget goes to the answer, which also keeps temperature valid.

Direct access to the manager

window.agentlet.ai.manager (also window.agentlet.aiManager) is the underlying AIManager instance, exposing the same methods plus getCurrentProvider().