LocalPilot

LocalPilot is a local Ollama-powered AI coding assistant for VS Code. It is built for developers who want inline suggestions, code explanations, fixes, tests, and chat without sending code to cloud AI services.

Features

• Copilot-style inline ghost-text suggestions powered by local Ollama models.
• Status bar menu for Ollama health, model selection, setup, chat, and inline mode.
• Chat panel for asking questions about selected code or the active file, with quick actions and code-block copy buttons.
• Selected-code commands for explaining code, adding comments, fixing code, generating tests, and solving coding problems.
• Code Actions for quick lightbulb access in common languages.
• Error explainer for selected stack traces or pasted terminal errors.
• Safe apply workflow with diff preview, copy option, and explicit apply confirmation.
• First-run setup that checks Ollama, lists local models, and asks before any model pull.
• Privacy-first defaults: no telemetry, no cloud AI APIs, and local-only model calls.

Screenshots

Screenshots will be added before Marketplace publication.

• Inline suggestions: resources/screenshots/inline-suggestions.png
• Chat sidebar: resources/screenshots/chat-sidebar.png
• Safe diff preview: resources/screenshots/diff-preview.png

Installation

Install LocalPilot from the VS Code Marketplace, or install a packaged VSIX:

code --install-extension localpilot-0.0.1.vsix

For local development:

npm install
npm run compile

Then press F5 in VS Code to launch the Extension Development Host.

Ollama Setup

Install Ollama from:

https://ollama.com/download

Start Ollama:

ollama serve

LocalPilot talks to this local host by default:

http://localhost:11434

You can change it with localpilot.ollamaHost.

Recommended Models

LocalPilot can guide setup on first run with LocalPilot: Run Setup. You can also pull models manually:

ollama pull qwen2.5-coder:1.5b
ollama pull qwen2.5-coder:7b
ollama pull smollm2:360m
ollama pull deepseek-coder:1.3b
ollama pull codegemma:2b

Suggested profiles:

• Balanced: qwen2.5-coder:1.5b for responsive local autocomplete and chat.
• Quality: qwen2.5-coder:7b for stronger local coding help on machines with more memory.
• Micro: smollm2:360m for very low-resource machines.
• Lite: deepseek-coder:1.3b for older low-end setups.
• Compact: codegemma:2b for a small general coding model.

LocalPilot never pulls a model automatically. It asks before downloads because model files can be large.

Low RAM Mode

Use localpilot.mode to control resource usage:

• auto: conservative defaults.
• micro: shortest context and smallest outputs.
• lite: balanced low-resource behavior.
• standard: more useful context and output sizes.
• custom: use your configured context and output settings.

Related settings:

• localpilot.inlineDebounceMs
• localpilot.disableInlineForLargeFiles
• localpilot.maxFileSizeKb
• localpilot.enableLowRamWarnings

Commands

• LocalPilot: Open Chat
• LocalPilot: Explain Selected Code
• LocalPilot: Explain Error
• LocalPilot: Add Comments to Selection
• LocalPilot: Fix Selected Code
• LocalPilot: Generate Tests
• LocalPilot: Solve Coding Problem
• LocalPilot: Check Ollama Status
• LocalPilot: Select Local Model
• LocalPilot: Switch Inline Completion Mode
• LocalPilot: Open Status Menu
• LocalPilot: Run Setup

Settings

• localpilot.ollamaHost
• localpilot.inlineModel
• localpilot.chatModel
• localpilot.lowRamModel
• localpilot.mode
• localpilot.enableInlineSuggestions
• localpilot.inlineCompletionMode (full by default, or line for conservative one-line suggestions)
• localpilot.maxContextLines
• localpilot.maxOutputTokens
• localpilot.temperature
• localpilot.inlineDebounceMs
• localpilot.disableInlineForLargeFiles
• localpilot.maxFileSizeKb
• localpilot.enableLowRamWarnings

Privacy

LocalPilot does not send code to cloud services. It talks to local Ollama by default and does not collect telemetry. See PRIVACY.md for details.

Troubleshooting

• Ollama disconnected: run ollama serve, then use LocalPilot: Check Ollama Status.
• No models found: run LocalPilot: Run Setup or pull a model with ollama pull <model>.
• Inline suggestions feel slow: switch localpilot.inlineCompletionMode to line, or switch localpilot.mode to micro or lite.
• Output looks weak: try a larger local model and set it with LocalPilot: Select Local Model.
• Files are skipped: LocalPilot blocks secrets, generated folders, lock files, and large files by design.

Known Limitations

• Model quality depends on the local model you install.
• Very small models may produce incomplete fixes or tests.
• LocalPilot does not read terminal contents automatically.
• Full inline completions are bounded and filtered, but local model quality still matters.
• Large workspaces and generated files are filtered aggressively for safety and responsiveness.

Development

npm run compile
npm run lint
npm test
npm run package