Zedge

Zedge brings AI-assisted coding to Zed through a local extension and a local companion sidecar.

The fair brag is architectural honesty: the extension is real, the sidecar is real, and the inference path is designed to stay close to the machine or network you control rather than disappearing into a generic hosted gateway.

The Two Parts

1. Zed extension in src/

   • Rust/WASM extension
   • registers Zedge as a language-model provider
   • adds slash commands to Zed

2. Companion sidecar in companion/

   • Bun HTTP server on localhost:7331
   • handles inference routing, collaboration bridges, compute pooling, and local integration work

Fast Path

The quickest way to try Zedge is to run only the companion and point Zed's OpenAI-compatible provider settings at it.

Start the companion

bun install
bun open-source/zedge/companion/src/index.ts

Then point Zed at:

• http://localhost:7331/v1

That gives you a working local provider path without compiling the extension first.

What People May Like

• the sidecar is OpenAI-compatible enough to plug into Zed quickly
• the extension adds native slash-command affordances when you want the full experience
• the companion is more than an inference proxy; it also handles collaboration and compute-pool responsibilities
• the architecture keeps local and edge-oriented routing options open

Building The Extension

cd open-source/zedge
cargo build --release --target wasm32-wasi

Then install it in Zed as a dev extension.

Companion Package

The companion is also its own package:

• @affectively/zedge-companion

It includes start, dev, build, and MCP entry points.

Why This README Is Grounded

Zedge does not need a giant architecture manifesto in the README. The strongest fair brag is that it already has both sides of the product: a Zed extension and a real local companion service.