Open-source building blocks
for the next era of AI retrieval.
Bicardinal is a small research lab shipping focused tools for multimodal search, indexing and reasoning. One install, one API — across PDFs, images, audio and text.
An open-source lab for AI retrieval.
Bicardinal is where research tooling meets production pragmatism. We design small, opinionated components that make it easy to ingest, embed, and search across every modality — without locking you into a black-box service.
Composable blocks
Every repository is a focused engine you can import, fork, or extend — not a monolithic platform.
Research-first
We build from the latest retrieval science and ship prototypes that graduate into production tools.
Open by default
Code, experiments, and roadmaps live on GitHub under Apache-2.0 so anyone can audit and contribute.
The essentials, nothing more.
A retrieval toolkit that stays out of your way — small surface, honest defaults, and every layer readable end to end.
One API, every modality
Text, PDFs, images and audio share a single index, a single query surface, and a single ranking model.
Local-first storage
Your corpus stays on disk. No hosted service, no telemetry, no data leaving your machine unless you ask.
Composable pipelines
Swap encoders, rerankers and vector backends with a few lines. Every layer is a small, readable Python module.
Production-ready
Streaming ingestion, incremental indexes, and typed results — designed to slot into real inference stacks.
A few lines of Python.
Install the package, point it at a folder, and query across every modality with a single typed call.
pip install bicardinalSmall tools. Sharp edges.
We don't believe in monolithic frameworks. Bicardinal is a collection of small, composable engines you can read in an afternoon, deploy in production, and bend to your own research questions.
Everything we ship is open source under Apache-2.0. No vendor lock-in, no hidden pipelines — just code, on GitHub, that does one thing well.
If you build with us, we'd love to see what you make.
Built in the open.
Read the source, file an issue, send a pull request. The repository is the product — every release lands there first.