`nervapack sync`¶

Incrementally update the graph after code changes — without re-ingesting the whole project.

Synopsis¶

nervapack sync [PATH]

Description¶

sync diffs your git working tree against the last commit, finds the files that changed, and surgically updates only those files in the graph and vector store.

Removes old nodes and ChromaDB vectors for each changed file.
Re-parses the updated file with tree-sitter.
Re-embeds the new entities into ChromaDB (batched in one call per file).
Re-binds any markdown docs linked to the changed file.
Saves the updated graph once at the end.

A full ingest on a large project can take minutes. sync turns that into a 2–5 second surgical update per file.

Requires git

sync uses GitPython to detect changed files. Your project must be a git repository.

Options¶

Argument	Description	Default
`PATH`	Path to the repository root	`.` (current directory)

Examples¶

# Sync after editing files
nervapack sync .

# Sync a different project
nervapack sync /path/to/project

Expected Output¶

Syncing changed files with NervaPack graph...
Found 3 changed files.
Updated AST for src/graph/builder.py
Updated AST for src/graph/retrieval.py
Removed old_module.py
Sync complete.

When to Use `sync` vs `ingest`¶

Situation	Command
First-time setup	`nervapack ingest .`
After editing 1–20 files	`nervapack sync .`
After a large refactor (many renames)	`nervapack ingest .`
After adding a new language/directory	`nervapack ingest .`
Graph corrupt or wrong data ingested	`nervapack clean --all && nervapack ingest .`

What Gets Updated¶

For each changed source file (.py, .js, .ts, etc.):

Old graph nodes for that file are removed (O(1) — file index lookup).
Old ChromaDB vectors for that file are deleted (delete_by_file).
The file is re-parsed by tree-sitter into new entities.
All new entities are batch-upserted into ChromaDB in one call.
New graph nodes and DEFINES edges are added.

For each changed markdown file (.md):

Old graph nodes removed.
Old vectors deleted.
File re-chunked by header.
Chunks upserted into ChromaDB.
LLM binding re-runs for the chunks (via LLMSummarizer).

nervapack sync¶