Skip to content
raph

Realtime sync

When an agent works on a codebase, raph keeps the graph updated in the background so lookups always reflect the latest code.

How it works

Section titled “How it works”

A filesystem watcher observes the indexed repositories. On a save it coalesces events with a ~150ms debounce, then incrementally re-indexes only the files that changed. Directories like .git, node_modules, vendor, and dist are ignored. A periodic reconcile catches anything missed and picks up newly registered repositories; if a watcher can’t be created, raph falls back to polling.

Commands

Section titled “Commands”
Terminal window
raph init --path .       # index and arm the watcher
raph sync --path .       # ensure the background worker is running
raph sync --status       # show worker status and registered repos
raph sync --stop         # stop the background worker
raph sync --remove --path .   # stop syncing a repo (and drop its graph data)

The worker starts automatically on raph init and when the MCP server starts, so in normal use you never manage it directly.