Licensing
What you can do with Docmancer code and TTAB-Bench.
One page covering the package and the evaluation data. If something is unclear, get in touch and we will document it here.
Python package: docmancer
The Docmancer Python package is published on PyPI under the MIT License. You can install, redistribute, and modify it for personal or commercial use without a separate agreement. The package source lives on GitHub; pull requests and issues are welcome.
Pipeline extras
The pipeline namespace (parsers, FTS5 store, Qdrant store, FastEmbed and hosted embedders, dispatcher, eval harness, and FastAPI server) ships under the same MIT terms as the base package. Optional integrations (Anthropic rerank, Voyage embeddings, OpenAI embeddings, Cohere embeddings) require a separate account with the respective provider and follow that provider’s terms.
TTAB-Bench evaluation dataset
TTAB-Bench (the labelled evaluation set used to score retrieval modes) is published on Hugging Face Datasets under CC BY 4.0. You can use it for research, benchmarking, or product development with attribution. If you publish numbers against it, include the dataset version (for example TTAB-Bench-v1.0) so comparisons stay honest.
Not legal advice
Nothing on this site is legal advice. Retrieval results and evaluation numbers published here are informational, not legal opinion or counsel.
Questions
For commercial questions, partnerships, or dataset queries, the project lead is reachable through the channels listed on the homepage.