Le Répertoire des AgentsFR · EN · ES · JA · HI

“I want agents that run at home, without sending my data to the cloud”

Professionals handling sensitive data (legal, health, finance), organizations subject to strict rules, and individuals who refuse to have their documents sent to American servers. You need a decent computer and some technical ease.

Recommended agent: Ollama (agents locaux)

💰 $0: the software and the open models are free; the real cost is your computer (a good graphics card makes the models fast and pleasant). Level: Advanced 🌍 United States Free, with no time limit; open models downloadable for free

Price verified on 2026-10-29.

Ollama (ollama.com), the reference tool for running open models (Llama, Mistral, Qwen…) directly on your computer. Plugged into an agent interface, it provides assistants that read and process your documents without a single line leaving your machine.

How it works

Trigger: You launch a task in your local agent interface — runs the model on your machine, processes your files locally and answers with no outside call, and you get 100% local answers and processing, traceable and free to use.

Step-by-step setup

Install Ollama on your computer (Windows, macOS or Linux).
Download an open model suited to your machine (a 7 to 14 billion parameter model is a good start).
Plug in a compatible agent interface (for example Open WebUI or AnythingLLM) to go beyond simple chat.
Test on real but non-critical documents, and compare quality with a cloud model before rolling it out.

⚠️ Guardrails: Local models are often a notch below the best cloud models in reasoning: keep critical tasks for human validation. And “local” doesn't mean magic: if you plug in a tool that sends data outside, privacy stops there.

🔧 Technical details

Installation: hybrid — local application (macOS, Windows, Linux, Docker); models hosted in Ollama's cloud optionally accessible from the local installation.
Prerequisites: macOS, Windows or Linux (or the official Docker image); no account required for local use, an Ollama account only for cloud models and model publishing.
Hardware: no global minimum configuration is published in the official documentation consulted; needs depend on the size of the loaded model (system memory or VRAM).
Self-hosting: yes — local binary or official ollama/ollama Docker image; server bound to 127.0.0.1:11434 by default, exposable via OLLAMA_HOST.
License: open_source — MIT License (MIT), LICENSE file in the official ollama/ollama repo.
API & integrations: API: yes — documented local REST API (generation, chat, model management), including OpenAI-compatible endpoints. · SDK: yes — official Python (ollama) and JavaScript (ollama-js) libraries. · CLI: yes — complete official CLI (run, pull, ps, stop, launch, etc.), published reference. · Webhooks: Not documented · MCP: Not documented
Models: Library of open models (Llama, Qwen, DeepSeek, Gemma, gpt-oss, Mistral, etc.) downloaded and run locally; custom models via Modelfile and GGUF/Safetensors import; Ollama cloud models optional. No third-party provider key required.
Data: Locally, Ollama states it doesn't see your prompts or your data; for cloud models, prompts and responses are processed to provide the service, without being stored or logged and without being used for training, according to the official FAQ. “Cloud” feature can be disabled (local-only mode).
Documented limits: Default context of 4,096 tokens (changeable via OLLAMA_CONTEXT_LENGTH or num_ctx); maximum queue of 512 requests by default (OLLAMA_MAX_QUEUE, 503 error beyond); number of simultaneously loaded models limited by available memory (default: 3 per GPU), according to the official FAQ.
Operations: Documented logs and troubleshooting; server configuration via environment variables (OLLAMA_HOST, OLLAMA_MODELS, OLLAMA_KEEP_ALIVE, etc.); models stored in ~/.ollama/models (location changeable). SSO and team roles: non_documente.

Official sources: ollama.com · docs.ollama.com · github.com — verified on November 1, 2026.

Find your agent with the quiz →

Frequently asked questions

How much does Ollama (local agents) cost?

$0: the software and the open models are free; the real cost is your computer (a good graphics card makes models fast and pleasant). Free, with no time limit; open models downloadable for free. Price verified on October 29, 2026.

Is Ollama (local agents) free?

Yes. Free, with no time limit; open models downloadable for free.

Is Ollama (local agents) hard to get started with?

Level: Advanced. Setup takes 4 steps; the first: Install Ollama on your computer (Windows, macOS or Linux).

What limits and guardrails should I know about with Ollama (local agents)?

Local models are often a notch below the best cloud models in reasoning: keep critical tasks for human validation. And “local” doesn't mean magic: if you plug in a tool that sends data out, privacy stops there.