Been annotating single-cell data for 7 years now. Got tired of second-guessing CellTypist/SingleR outputs in disease datasets. Reference methods pattern-match against atlases. Works until it doesn't: disease, ectopic expression, unexpected tissue context. You get an annotation label and no way to know it's wrong.

CyteType runs multiple LLM agents that propose and critique annotations and surface all the ambiguity instead of hiding it.

The output is an interactive report you can chat with to interrogate the reasoning. Linked cell ontology terms and relevant literature for each cluster. Confidence score and match score against author labels help with triaging.

Model-agnostic. Benchmarked across 16 LLMs and against reference-based methods. Integrates with Seurat/Scanpy/Anndata.