Home > FAQs > vs GitHub Copilot

OpenCues vs GitHub Copilot: what is the difference?

GitHub Copilot autocompletes code inside a code editor. OpenCues is a different tool for a different surface: it adds AI to any text input (browser fields, agentic CLI prompts, your terminal), and it inverts who chooses the moment.

The key differences

Where it lives. Copilot's completions live in code editors and target source code (its chat, CLI, and agent surfaces are a separate story). OpenCues lives wherever you type prose or prompts, in any supported host.
Who initiates. Copilot's ghost text appears when the model decides. With OpenCues you place a _ exactly where you want the response; the system chooses only what to put there, never where or when.
What it can touch. Beyond generation, OpenCues offers word alternatives (cues), continuous background rewriting (agent tasks), and blanks that read or set live system state (volume _, weather _).
Model choice. Copilot picks from GitHub-hosted models, with bring-your-own-key on some surfaces. OpenCues is model agnostic everywhere: Cerebras, Groq, OpenAI, Anthropic, Gemini, OpenRouter, or local models via Ollama.
Openness. OpenCues is free, open source, and defined by an open standard anyone can implement.

They are complementary: Copilot writes code in your editor; OpenCues improves the prompts and prose around your work, including the prompt you type into an AI coding agent.

Smaller differences that matter day to day: OpenCues edits land dimmed and revertable per word, each feature routes to its own model (Copilot's model picker covers chat), and project-level cues committed to your repo play the role of Copilot's per-repo instructions, for prose instead of code.

Where Copilot struggles

Its world ends at the IDE: the ghost text that defines it exists only in code buffers, and completions cannot run locally or on your own key (chat can, completions cannot). Acceptance is the only gesture: once ghost text lands it is plain code, and unwinding it is the editor's undo, not a managed span.

By the numbers

In its own arena Copilot is fast: type code, Tab, two steps with one of them the accept. Outside code the numbers diverge: five steps for an inline-chat rewrite against OpenCues' one, and an accepted suggestion unwinds only through editor undo, whole, where OpenCues reverts per word. Sequences in the full comparison.