SMF·GITHUB-COPILOT
Can a prompt replace GitHub Copilot?
Developer tools — AI coding assistant
Exhibit tracking slip
Verdict
The individual pieces — an open-source in-editor assistant, a model API, a local completion model — are all real and available today. Assembling them into a working setup is a weekend project. What doesn't carry over is Copilot's own low-latency completions tuned specifically for the editor, and its GitHub-native context.
Exhibit A — The prompt
Received on30.07.2026Do not build or train a model — set up the open-source Continue extension in VS Code (or your editor of choice), configured against models you control. Use a capable model over API for chat and inline edit (key read from an environment variable, never committed), and a small local model served by Ollama for autocomplete, so completions are free and keep working offline, at some cost in speed and quality versus Copilot's own. Deliver a working Continue config with both models wired up and sensible keybindings for inline edit and chat. Turn on Continue's codebase indexing so answers are repo-aware, and document where the index lives and how to rebuild it. Add a small script that pipes `git diff` to the chat model and prints a short pre-commit review, as a rough stand-in for Copilot's PR-review features. Turn off the extension's own telemetry in its config.
Do not build a full editor extension from scratch, GitHub PR-bot automation, or team seat management — those are real platform features, not something to reassemble in a weekend. Note in the README that local autocomplete is slower and less accurate than Copilot's; that gap is the trade for free and private completions.Opening prefills the prompt — press enter to run it.
Exhibit B — What you lose
- B.1 native low-latency completions tuned for the editor
- B.2 GitHub-native context and PR features
- B.3 model routing handled for you
- B.4 managed billing and team controls
Prior art
Exhibit C — Why people still pay: model/platform/editor integration
Fast, repo-aware completions are available in the editor with defaults that already work — people pay to skip the setup, not because the underlying capability is unavailable elsewhere.
Questions
Can I bring over my Copilot settings or custom instructions?
Not directly — Copilot's configuration lives on GitHub's side. Continue reads its own config file, so any custom instructions or preferences need to be re-entered there by hand.
Will it work on a laptop with no internet?
Chat and inline edit still need a connection to reach the model API. The local Ollama-served autocomplete is the one part that works fully offline, which Copilot itself doesn't offer at all.
What does it actually cost to run?
Only API usage for chat and inline edit — Continue and Ollama are both free. Light-to-medium use is often cheaper per month than a flat Copilot seat; heavy daily use can end up costing more.
What's the one thing that genuinely doesn't survive the rebuild?
Speed. Copilot's completions are tuned end-to-end for low latency inside the editor; a local model through Ollama is noticeably slower and less accurate, which you feel on every keystroke.
Related tools
Receipt