triplum

A Python library being rebuilt for composable document search, knowledge graphs and GraphRAG experiments. Combine small functions and classes to compare pipelines and their baselines.

Try the building blocks

From a repository checkout, run uv sync to install (Python 3.14+), then uv run python to try this example:

from triplum.datatype import Source
from triplum.steps.chunking import FixedSize

source = Source(origin="example", text="Alpha. Beta.")
for chunk in FixedSize(7)(source):
    print(chunk.start, repr(chunk.text))

0 'Alpha. '
7 'Beta.'

Each chunk retains its source reference and character offset. Follow chunking to replace this naive splitter, or Core concepts for the full sequence of building blocks. For intermediate-result reuse, see the cache guide.

Project status

Indexing and retrieval distinguishes implemented foundations from the intended flow. Indexing prototypes remain under review; there is no supported end-to-end retrieval pipeline. The Python package and Rust workspace are independent, so the example requires no Rust build.