🛫 The Local–Cloud Boundary
Copilot is not one thing running in one place. Every completion, chat reply, and agent action crosses a boundary between your machine and GitHub's cloud. The Runway maps that boundary: where context is gathered, where inference happens, where results land — and the five distinct data flows that connect them.
⬡ Handover Topology
Where Copilot runs and how requests flow between your machine and the cloud. Every completion, chat message, and agent action crosses at least one of these boundaries.
Your IDE extension host — VS Code, JetBrains, Neovim, CLI. Gathers context (open files, repo index, terminal), sends requests, renders results. All prompt construction happens here.
Frontier models hosted by GitHub (proxied to OpenAI, Anthropic, Google, xAI). Your prompts leave the IDE, hit GitHub's API, get routed to the provider, and the completion streams back.
GitHub-hosted headless agents — Cloud Agent, Copilot Workspace. No local IDE involved: the entire loop (read repo → plan → code → test → PR) runs on GitHub infrastructure.
Bring Your Own Key routes traffic through your Azure OpenAI, Anthropic, or Google Cloud tenant instead of GitHub's proxy. You control retention, billing, and regional pinning — at the cost of setup complexity.
Data Flows
⚠ NOTAMs
Notices to Air Missions — deprecations, preview status caveats, regional availability, anything that changes routing decisions.
🛠 Copilot Engine
flowchart LR
U[You - IDE / CLI / GitHub.com] --> E[Copilot Engine]
E --> R[Router]
R --> P[Policy Gate]
P --> X[Context Assembler]
X --> M{Model Pool}
M --> O[OpenAI]
M --> A[Anthropic]
M --> G[Google]
M --> I[xAI]
M --> F[GitHub fine-tunes]
M --> RES[Response]
RES --> AUD[Audit Log]
AUD --> U
◉ Departure Board
Full model reference. Filter by plan, provider, or status — click any row for the complete briefing.