All projects

Graph Index

Tags every file from a closed vocabulary and answers tag queries offline

In progress

Stack
GoTask

Gives every file in a folder a weighted set of tags from a closed English vocabulary of 100k, 1M or 10M entries, code tags such as lang:pl and year:2024 included. A tag query returns the matching files. It never invents a tag, and nothing leaves the machine.

How it works

  • Tags come only from the closed vocabulary, each with a weight and the evidence behind it.
  • Queries combine AND, OR, NOT, phrases, year ranges and weight filters, from the command line or over MCP.
  • No network calls at runtime. A local LLM summary pass exists and is off by default.

Example

# Index a folder, then ask which files carry the tags
task indexer:run -- /abs/path/to/folder
cd indexer
go run ./cmd/graph-index tags query /abs/path/to/folder "invoice AND year:2024"