drsg-embedded

The dr-strange graph engine inside your own process: no server, no socket, no port. One file on disk, opened directly.

import asyncio
import drsg_embedded as drsg

async def main():
    db = await drsg.open("graph.drsg")
    print(await db.planes())
    print(await db.context("myrepo", "Service.Save"))
    print(await db.impact("myrepo", "normalize", depth=3))
    print(await db.deployed("myrepo"))

asyncio.run(main())

Every method is a coroutine, and every one does its work on a blocking thread rather than on the event loop. A point lookup here is microseconds of CPU, not I/O to wait on, so running it inline would stall every other task for exactly as long as it takes.

Calls name their plane. A plane handle borrows the database it came from and a Python object cannot hold a borrow, so the surface is the one the MCP server and the CLI already use — db.context("myrepo", "Service.Save").

What this is and is not

This wraps the engine: planes, nodes, edges, and the agent verbs over a graph that already exists. It does not digest a repository — that needs the sandboxed wasm parser plugins, which would put tens of megabytes into every wheel. Build a plane with the CLI and read it from here:

$ drsg digest ./myrepo --apply --plane myrepo

For talking to a running drsg serve over JSON-RPC instead, the pure-Python client is drsg.

Status

Alpha. The surface above is settled; cypher, the change feed and bulk writes are the next additions.