LocalPilot

Getting started

Install LocalPilot, connect a local model server and run your first agent task in under five minutes.

Ask AI about this page
ChatGPT Claude Perplexity Gemini View as Markdown

1. Serve a model — or let LocalPilot do it

Nothing installed yet? Skip this step. When LocalPilot starts without finding a model it offers Set up automatically: it installs Ollama in the background (official build, no admin rights), starts it and downloads a starter coder model sized for your RAM and GPU — a 7b on 16 GB+, smaller on tighter machines — plus a tiny completion model. One click, a few minutes of download, done. Details in Automatic setup.

Already have a server? LocalPilot talks to any of these out of the box (default ports are probed automatically):

Server Default endpoint Notes
Ollama http://localhost:11434 Native API: tool calling, thinking, FIM, vision
LM Studio http://localhost:1234/v1 Start the server from the Developer tab
llama.cpp llama-server http://localhost:8080/v1 Great for Unsloth GGUF exports
vLLM / SGLang http://localhost:8000/v1 GPU servers, OpenAI-compatible 
# Ollama example — Gemma 4 e4b: tools + vision + thinking in ~6.6 GB
ollama pull gemma4:e4b
ollama serve

Any other OpenAI-compatible server (LocalAI, Jan, text-generation-webui, OpenRouter, …) works too — add it under localpilot.providers. See Models & providers.

2. Install the extension

From the Marketplace (search for LocalPilot Agent in the Extensions view), or from the command line:

code --install-extension localpilot.localpilot-agent

From a GitHub release: download the latest VSIX from the releases page (or the direct link) and run code --install-extension localpilot.vsix, or use Extensions view → ··· → Install from VSIX…

3. Open LocalPilot

Click the LocalPilot icon in the Activity Bar or press ⌘⇧L (Ctrl⇧L on Windows/Linux).

On first launch the Welcome / Setup screen opens: it detects the servers from step 1, lists their models with a recommended default, and lets you choose how much freedom the agent gets (Careful, Balanced or Autonomous). Click Finish & open chat and you’re done. You can rerun it any time with LocalPilot: Welcome / Setup.

Setup step 1 inside VS Code — LocalPilot found a running Ollama with 2 models

Setup step 2 — choosing a default model; qwen2.5-coder:14b is recommended because it supports tool calling

If no server is found the wizard shows one button for your situation — Install Ollama & starter model, Start Ollama & download starter model or Download starter model — with the models and download size it will fetch. One click downloads and installs Ollama in the background — no admin rights, no terminal — starts it, pulls the starter models with live progress, and moves you on to picking the default model.

Afterwards, pick a model from the picker at the bottom of the composer — or click the model name in the status bar. If you see No models found, click Set up automatically, or make sure a server is running and press Refresh. Settings → Models shows what fits your machine and lets you install more from the catalog; see Managing models.

4. Your first task

Switch to Agent mode and try:

Add input validation to the login form and write tests for it.

LocalPilot will read the relevant files, make edits, run diagnostics and report back. Review the files changed bar above the composer and choose Keep all or Undo all.

Next steps