Ollama

O que é o Ollama?

Ollama was launched in late 2023 by the team behind the popular Llama model. It ships as a simple command‑line tool that downloads model binaries, stores them locally, and exposes a REST API for inference. The project has received regular updates, with the latest release adding support for GPU acceleration on both NVIDIA and AMD hardware.

The target audience is developers, researchers, and power users who want to experiment with LLMs on their own machines. It is especially useful for those who prefer not to rely on cloud services due to privacy or bandwidth concerns.

Compared to Windows‑centric AI assistants like Microsoft Copilot, Ollama feels more hands‑on. You control the entire stack, from model selection to inference speed, and you can integrate it into scripts or IDEs without waiting for a vendor’s update cycle.

Ideal para: Users who need full control over LLMs and prefer local inference over cloud services.

Principais recursos do Ollama

• Local model hosting via Docker or native binaries
• Supports Llama, Mistral, and other open‑source models
• GPU acceleration with CUDA, ROCm, or CPU fallback
• REST API for easy integration
• CLI for model download, list, and removal
• Built‑in model quantization for memory savings
• Cross‑platform packaging (Arch, openSUSE, Fedora)

Dicas para quem vem do Windows

If you’re coming from Microsoft Copilot, note that Ollama does not provide a graphical assistant out of the box. You’ll interact via the command line or by calling its REST API from scripts. Model files are stored in ~/.ollama/models, and you can switch models with ollama list and ollama use <name>. Keyboard shortcuts are irrelevant; instead, focus on API endpoints and CLI flags. Workflow changes: instead of a single integrated assistant, you’ll run a local server and send prompts programmatically.