Context for the agent,without the upload.
Retrieval usually means shipping your code to something. This parses it where it already is and leaves the result there.
Retrieval usually means upload.
To give an assistant context, most tools index your repository somewhere else: chunks through an embedding API, a hosted service holding a copy, a platform to deploy and then to audit. Each one is a new place your source lives.
For a lot of organisations that ends the conversation, and the result is teams using assistants with no codebase context at all rather than with the wrong kind.
Where each part runs.
Structurally different rather than configured differently: there is no remote half to turn off.
- Parsing
- Tree-sitter, on your machine. Thirty-six grammars, no network call, nothing to opt out of.
- Storage
- Three files in your own repository under graphify-out. There is no backend to hold a copy.
- Telemetry
- None. Not off by default: absent, and the source is Apache 2.0 so you can confirm it.
One caveat, stated plainly. Graphify adds no new place your code travels, and it does not change where your assistant sends its prompts. If your assistant calls a hosted model today, it still does, under the terms you already accepted for it.
The deeper passes that read prose and name communities run through that same session, or through a local backend you point them at. The structural graph is pure syntax work and needs no key at all.
For air-gapped work, a local model backend covers the inferred edges and the parser covers the rest. For a self-hosted layer with verification and graph-aware review, that is Enterprise, in early access.