# Local AI inference: your data never leaves

A sovereign LLM server. On our GPUs in France or deployed on your infrastructure. Predictable costs, zero US dependency.

## What it does

A self-hosted LLM inference server that runs the best open-source models on our GPUs in France.

### Self-hosted

Runs on our GPUs in France or directly on your infrastructure. No external dependency.

### Integrated REST API

Inference is exposed via the Agora API. OpenAI-compatible endpoints are also available. Your developers work with a format they already know.

### Multi-model

Qwen, Mistral, Llama, DeepSeek. We deploy the best-performing models for your needs.

### Open-weight

Qwen, Mistral, Llama, DeepSeek. The best open-weight models. No lock-in with a proprietary vendor.

## Why it's different

### Predictable costs, no tokens

Monthly flat rate included in the subscription. No token counter, no nasty surprises.

### Your data stays with you

On our servers in France or directly on your infrastructure, your choice.

### Zero US cloud dependency

No dependency on OpenAI, AWS or Google. Your AI stays autonomous.

## Reasoning budget

A unique feature: finely control the "thinking time" allocated to each task.

- Simple task = minimal budget, instant response
- Complex task = high budget, in-depth reasoning
- Automatic cost/quality optimization

Simple questionLow budget

Document analysisMedium budget

Complex generationHigh budget

## Multi-model orchestration

The right model for the right task. Orchestration is automatic.

### Fast model

Small model for simple tasks: classification, extraction, factual answers. Response in milliseconds.

### Powerful model

Large model for complex generation: long-form writing, multi-step reasoning, fine-grained analysis.

## Vision and OCR

Multimodal models understand images, scanned documents and photos. Extract information from any visual medium.

### Images

Photos, screenshots, diagrams. The model describes and analyzes visual content.

### Scanned documents

Invoices, contracts, forms. Built-in OCR to extract text and structured data.

### Field photos

Construction site photos, parts, labels. Visual understanding for business use cases.

## AI Recipes

Pre-defined configurations by use case. System prompt, parameters, guaranteed output format. Ready to use.

### Summary

Automatic synthesis of long texts

### Extraction

Structured data from free text

### Classification

Automatic sorting of tickets, emails, documents

### Writing

Controlled business content generation

### Translation

Contextual multilingual translation

### Q&A

Precise answers from a document base

## Flexible deployment

Our inference stack can be deployed on your infrastructure, on your servers, your GPUs.

- **On-premise LLM** : inference runs on your GPUs, data never leaves your network
- **Embedded vector database** : RAG works locally, documents and embeddings stay with you
- **Deployment and maintenance** : we deploy, configure and maintain the stack over time

Agora Cloud

Our GPUs in France

or

On your premises

Your GPUs, your private cloud

## Take back control of your AI

Let's assess your infrastructure and inference needs together.

[Request a technical audit](https://www.agora.software/en/contact/)