Brings engined into VS Code's own Copilot Chat surface: engined's local models sit in the model picker beside Copilot's, its image/vision/audio/search routes attach as ordinary chat tools, and its completions route offers inline ghost text. Everything routes to engined running on the same machine, or reached over VS Code Remote-SSH.
The extension only ever talks to engined.doors; no request goes anywhere else, and the status bar always shows which route answered and whether it stayed local.
Then open Copilot Chat's model picker and choose an engined model. Full runbook (building from source, attaching a tool, verifying, uninstalling): HUMANS.md.
Features
Local chat models in Copilot's model picker, streamed through engined
Five agent tools: generate/edit images, OCR or describe an image, transcribe or translate audio, synthesize speech, semantic workspace search
Inline completions (ghost text), optionally with neighbouring-file context
Status bar shows what answered, whether it ran locally, tokens used, and a loading state
An Engines view for warming, holding, and stopping engines, with live resource usage
A usage report (per-day/per-route requests, tokens, cost, and local/remote split) across every configured door
Image attachments work on a text-only engined model that engined bridges to a vision route
Copilot's Context Window meter shows the token usage engined reports for each reply
Screenshots
Fill-in-the-middle completions from a local model
Status popup: what answered, where, and at what cost
Engines view with live resource use
engined's tools in Copilot's tool picker
Speech-to-text through #enginedTranscribe
Usage report across routes
Copilot's Context Window meter with engined's counts
