ai-mesh

Thin wrapper around Vercel AI SDK with DI support. Use any provider the SDK supports.

Install

npm install @ekipnico/ai-mesh

Quick Start

import { Mesh } from 'mesh-ioc';
import { LlmService, LlmProvider, VercelLlmProvider, ModelRegistry, UsageTracker } from '@ekipnico/ai-mesh';

const mesh = new Mesh('MyApp');
mesh.service(LlmService);
mesh.service(ModelRegistry);
mesh.service(UsageTracker);
mesh.service(LlmProvider, VercelLlmProvider);

const llm = mesh.resolve(LlmService);

const response = await llm.generate({
    model: 'gpt-4o',
    prompt: 'Hello!',
});

Using Any Provider

Pass any Vercel AI SDK model directly:

// Groq
import { createGroq } from '@ai-sdk/groq';
const groq = createGroq({ apiKey: process.env.GROQ_API_KEY });

await llm.generate({
    model: groq('llama-3.1-70b-versatile'),
    prompt: 'Hello!',
});

// Mistral
import { createMistral } from '@ai-sdk/mistral';
const mistral = createMistral({ apiKey: process.env.MISTRAL_API_KEY });

await llm.generate({
    model: mistral('mistral-large-latest'),
    prompt: 'Hello!',
});

// DeepSeek
import { createDeepSeek } from '@ai-sdk/deepseek';
const deepseek = createDeepSeek({ apiKey: process.env.DEEPSEEK_API_KEY });

await llm.generate({
    model: deepseek('deepseek-chat'),
    prompt: 'Hello!',
});

// Any other @ai-sdk/* provider works the same way

Built-in Providers

For OpenAI, Anthropic, and Google, just use string IDs:

# Set API keys
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=AIza...

await llm.generate({ model: 'gpt-4o', prompt: '...' });
await llm.generate({ model: 'claude-sonnet', prompt: '...' });
await llm.generate({ model: 'gemini-flash', prompt: '...' });

New models auto-detected by name pattern:

await llm.generate({ model: 'gpt-5', prompt: '...' });           // → openai
await llm.generate({ model: 'claude-opus-4.5', prompt: '...' }); // → anthropic
await llm.generate({ model: 'gemini-3-flash', prompt: '...' });  // → google

API

generate(params)

const response = await llm.generate({
    model: 'gpt-4o',             // string ID or LanguageModel
    context: 'You are helpful.', // system prompt
    prompt: 'Hello',             // required
    messages: [],                // conversation history
    images: [],                  // for multimodal
    maxTokens: 1000,
    temperature: 0.7,
});
// → { content, usage, finishReason }

generateObject(params)

const person = await llm.generateObject({
    model: 'gpt-4o',
    prompt: 'Extract: John is 30',
    outputSchema: PersonSchema,
});
// → parsed object

stream(params)

for await (const chunk of llm.stream({ prompt: 'Tell a story' })) {
    if (chunk.type === 'text') process.stdout.write(chunk.content);
}

Registered Models

| ID | Provider | Actual Model | |----|----------|--------------| | gpt-4o | openai | gpt-4o | | gpt-4o-mini | openai | gpt-4o-mini | | gpt-4.1 | openai | gpt-4.1 | | gpt-4.1-mini | openai | gpt-4.1-mini | | gpt-4.1-nano | openai | gpt-4.1-nano | | o3 | openai | o3 | | o4-mini | openai | o4-mini | | claude-sonnet | anthropic | claude-sonnet-4-20250514 | | claude-sonnet-4.5 | anthropic | claude-sonnet-4-5-20250929 | | claude-opus | anthropic | claude-opus-4-20250514 | | claude-opus-4.5 | anthropic | claude-opus-4-5-20251124 | | claude-haiku | anthropic | claude-haiku-4-5-20251215 | | gemini-flash | google | gemini-2.5-flash | | gemini-flash-lite | google | gemini-2.5-flash-lite | | gemini-pro | google | gemini-2.5-pro | | gemini-2.0-flash | google | gemini-2.0-flash |

Usage Tracking

const stats = llm.getUsage();
console.log(stats.tokens);   // { input, output, total }
console.log(stats.cost);     // estimated USD (registered models only)
console.log(stats.requests);

llm.resetUsage();

DI Pattern

export class MyAgent {
    @dep() private llm!: LlmService;

    async run() {
        return await this.llm.generate({ prompt: 'Hello' });
    }
}