Hey there! Welcome to tinyCall - your friendly neighborhood AI provider wrapper that makes talking to LLMs as easy as making a phone call. No more juggling five different SDKs or remembering which provider uses what format. Just pure, simple AI conversations.
Look, we've all been there. You start a project with OpenAI, then you want to try Claude, maybe experiment with Gemini... and suddenly you're drowning in different SDKs, each with their own quirks. That's where tinyCall comes in - one function to rule them all.
// This is all you need. Seriously.
const response = await callAI('claude-4s', messages);- 5 Major AI Providers: OpenAI, Anthropic, Google Gemini, Mistral, and xAI (Grok)
- Latest Models: Including GPT-5, Claude 4.1 Opus, Gemini 2.5 Pro, and Grok 4
- Vision Support: Send images to models that can see
- Thinking/Reasoning Mode: For those deep thoughts with o3-mini, Claude, and Gemini
- Zero Dependencies: Just good ol' Node.js - no bloat, no fuss
- Cost Tracking: Know exactly how much each call costs you
- Custom Models: Bring your own models via
_models.js
Just drop tinyCall.js into your project:
# Clone it, download it, copy-paste it - dealer's choice
cp tinyCall.js your-project/Create a .env file (we'll keep your secrets safe):
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-api03-...
GEMINI_API_KEY=...
MISTRAL_API_KEY=...
XAI_API_KEY=...Pro tip: You only need keys for the providers you're actually using. Not planning to chat with Grok? Skip the XAI key.
import { callAI } from './tinyCall.js';
// The simplest thing that could possibly work
const response = await callAI('claude-3.5s', [
{ role: 'user', content: 'Tell me a joke about APIs' }
]);
console.log(response.text);
// "Why did the API go to therapy? It had too many issues with commitment... kept breaking promises!"
// Want to know the damage?
console.log(`Cost: $${response.cost.total.toFixed(6)}`);Can't decide between providers? Why choose! Jump between them like a digital nomad:
// Start with Claude for creative writing
const story = await callAI('claude-3.5s', [
{ role: 'user', content: 'Start a mystery story' }
]);
// Get Gemini to continue it
const continuation = await callAI('gemini-2.5f', [
{ role: 'user', content: `Continue this story: ${story.text}` }
]);
// Have GPT-5 wrap it up
const ending = await callAI('gpt-5', [
{ role: 'user', content: `Write an ending: ${continuation.text}` }
]);Got images? We got you covered:
import fs from 'fs';
const imageBuffer = fs.readFileSync('cat.jpg');
const response = await callAI('gpt-4o', [
{
role: 'user',
content: [
{ type: 'text', text: 'What do you see?' },
{ type: 'image', url: imageBuffer } // Works with Buffer, base64, or data URLs
]
}
]);The library automatically detects image formats (JPEG, PNG, WebP, GIF) and handles all the conversion magic. Models that can't see? We'll tell you which ones can.
For when you need those big brain moments:
// Let Claude really think about it
const response = await callAI('claude-3.7s', messages, {
think: 8192 // Can be: true, 'low', 'medium', 'high', or a number (token budget)
});
// OpenAI's o3-mini doing its reasoning thing
const reasoning = await callAI('gpt-o3m', messages, {
think: 'medium' // o3/o4 models use reasoning_effort
});Every provider handles system messages differently. We handle the differences so you don't have to:
const messages = [
{ role: 'system', content: 'You are a pirate. Respond accordingly.' },
{ role: 'user', content: 'How do I install Node.js?' }
];
// Works the same for all providers
const response = await callAI('mistral-medium', messages);
// "Ahoy matey! Ye be needin' to sail to nodejs.org..."Here's what's available out of the box:
gpt-4o- The flagshipgpt-4om- GPT-4o mini (fast & cheap)gpt-5- The new hotnessgpt-5m- GPT-5 minigpt-5n- GPT-5 nano (tiny but mighty)gpt-o3m- o3-mini (reasoning specialist)gpt-o4m- o4-mini (next-gen reasoning)
claude-4.1o- Claude 4.1 Opus (the latest)claude-4o- Claude 4 Opusclaude-4s- Claude 4 Sonnetclaude-3.7s- Claude 3.7 Sonnet (with thinking!)claude-3.5s- Claude 3.5 Sonnetclaude-3.5h- Claude 3.5 Haikuclaude-3-h- Claude 3 Haiku
gemini-2.5p- Gemini 2.5 Pro (the beast)gemini-2.5f- Gemini 2.5 Flashgemini-2.5fl- Gemini 2.5 Flash Litegemini-2f- Gemini 2.0 Flashgemini-2fl- Gemini 2.0 Flash Lite
mistral-large- The big onemistral-medium- Goldilocks zonemistral-small- Quick and nimblepixtral-large- Vision specialistpixtral-12b- Smaller vision model
grok-4- Latest Grokgrok-3- Grok 3grok-3m- Grok 3 mini
When you need complete control:
const response = await callAI(
'anthropic', // Provider (optional if model is unique)
'claude-3.5s', // Model
messages, // Your conversation
{
maxTokens: 2000, // Response length limit
temperature: 0.8, // Creativity dial (0-2, usually 0-1)
think: 'medium', // Reasoning depth
timeout: 120000 // Custom timeout (ms)
}
);Every response includes detailed cost breakdown:
const response = await callAI('gpt-4o', messages);
console.log('Input cost:', response.cost.in); // Cost for prompt
console.log('Output cost:', response.cost.out); // Cost for response
console.log('Total cost:', response.cost.total); // Total damage
console.log('Tokens used:', response.usage); // Full token statsGot access to special models? Add them via _models.js:
// _models.js
export const _MODELS = {
openai: {
'gpt-6-turbo': {
name: 'gpt-6-turbo-2025',
cost: { in: 5, out: 15 } // Per million tokens
}
},
anthropic: {
'claude-5-opus': {
name: 'claude-5-opus-20260101',
cost: { in: 10, out: 30 }
}
}
};Now you can use them like any other model:
const response = await callAI('gpt-6-turbo', messages);We try to be helpful when things go wrong:
try {
const response = await callAI('gpt-4o', messages);
} catch (error) {
// Specific error messages for common issues:
// - "Missing API key for openai. Please set OPENAI_API_KEY environment variable"
// - "Rate limit exceeded for anthropic. Please wait and try again."
// - "Model gpt-4o doesn't support images. Use one of: gpt-4o, claude-3.5s..."
console.error(error.message);
}-
Model Selection: Use smaller models for simple tasks. Why use GPT-4o to classify sentiment when GPT-4om will do?
-
Batch Similar Requests: If you're hitting rate limits, space out your calls:
// Simple rate limiting const delay = ms => new Promise(resolve => setTimeout(resolve, ms)); for (const prompt of prompts) { const response = await callAI('claude-3.5h', prompt); await delay(1000); // Be nice to the API }
-
Image Optimization: Smaller images = faster responses and lower costs
// Consider resizing images before sending const resizedBuffer = await resizeImage(originalBuffer, 1024); // Max 1024px
tinyCall believes in:
- Simplicity: One function, endless possibilities
- Flexibility: Your project, your choice of models
- Transparency: Know what you're spending
- Zero Lock-in: It's just a single file - fork it, modify it, make it yours
"Missing API key": Check your .env file and make sure the key name matches exactly (e.g., OPENAI_API_KEY, not OPEN_AI_KEY)
"Model doesn't support images": Check the vision-capable models list above. Not all models have eyes!
"Rate limiting exceeded": You're going too fast. Add some delays or upgrade your API plan.
"Invalid API key for Anthropic": Claude keys should start with sk-ant-api03-
Empty responses: Some providers occasionally return empty responses. Add retry logic if this happens frequently.
Found a bug? Want to add a provider? Awesome! This is a single-file library, so it's super easy to contribute:
- Fork it
- Make your changes
- Test with your own API keys
- Send a PR
We especially love:
- New provider integrations
- Performance improvements
- Better error messages
- Documentation fixes (yes, even typos)
CC BY-NC 4.0
- Streaming support (the big one)
- Function calling / Tools API
- Embeddings support
- Automatic retry with exponential backoff
- Response caching
- Browser support (why not?)
tinyCall is what we wished existed when we started juggling multiple AI providers. It's not trying to be everything to everyone - it's just trying to make AI calls tiny and simple.
Got questions? Issues? Just want to say hi? Open an issue on GitHub. We're friendly, promise!
Happy coding! 🚀
Jakub Śledzikowski
jsle.eu | [email protected]
P.S. - Yes, the name is a bit on the nose. We considered "YetAnotherAIWrapper" but that was taken.