Skip to content

zed.md

Latest commit

 

History

History
56 lines (39 loc) · 1.71 KB

zed.md

File metadata and controls

56 lines (39 loc) · 1.71 KB

engram — Zed Context Server

Exposes engram's knowledge graph as a Zed slash command via the context server protocol (JSON-RPC over stdio).

Prerequisites

  • engram installed globally: npm install -g engramx
  • Project indexed: engram init run in your project root

Setup

Add to your Zed settings.json:

{
  "context_servers": {
    "engram": {
      "command": {
        "path": "engram",
        "args": ["context-server"]
      }
    }
  }
}

Open ~/.config/zed/settings.json (macOS) or ~/.local/config/zed/settings.json (Linux) and merge the block above.

Usage

In Zed's agent panel, type /engram followed by your query:

/engram auth flow
/engram database schema decisions
/engram known issues GraphStore

engram queries the local knowledge graph and injects matching context — architecture nodes, past decisions, mistake warnings — directly into the AI's prompt.

How It Works

  1. Zed sends context/list on startup — engram advertises the engram slash command.
  2. When you invoke /engram <query>, Zed sends context/fetch with { query, project }.
  3. The server runs engram query <query> -p <project> --budget 2000 as a subprocess.
  4. The result (nodes, edges, mistake warnings) is returned as text and injected into context.

Optional: Per-Project Config

Pass a specific project path via the project param. By default the server uses process.cwd() at the time Zed launches it.

Troubleshooting

  • "engram query failed" — run engram init in your project root, then engram (mines the codebase).
  • No results — try a broader query, or run engram stats -p . to check node count.
  • Command not found — ensure engram is on your PATH (which engram).