GitHub Copilot

Build me a local-first coding assistant setup to replace GitHub Copilot. Requirements: - Do not build or train a model. Set up the open-source Continue extension in VS Code, configured against models I control. - Chat and edit model: Claude or GPT over API, key referenced from my environment, never committed. Autocomplete model: a small local coder model served by Ollama, so completions are free and work offline. - Deliver the actual Continue config file with both models wired, plus sensible keybindings for inline edit and chat. - Enable Continue's codebase indexing for repo-aware answers; document where the index lives and how to rebuild it. - Add scripts/ai-review.sh: pipes `git diff` to the API model (key from .env) and prints a short pre-commit review. - Everything local except the chat-model API calls; turn off the extension's telemetry in the config. - Out of scope: writing an editor extension from scratch, GitHub PR-bot features, team seat management. - README: Ollama install and model pull commands, where keys go, and an honest note that local autocomplete is slower and dumber than Copilot's, that is the trade for free and private.