Why AI coding context should start local.

AI coding tools can be useful without turning every context decision into a cloud platform decision. A local-first context layer starts from the developer workspace, writes local artifacts, and keeps packet review close to the code.

Local first does not mean isolated.

The packet can still be copied into GitHub Copilot, served through MCP, or reviewed in VS Code. The important part is that packet construction happens where the source, specs, tests, exclusions, and team instructions already live.

Metrics without raw source

Teams need to understand context quality: budget pressure, repeated lookups, stale specs, redactions, and estimated tokens saved. Those metrics can be useful without collecting raw code by default.