We've closed a $61M Series B led by Thicket Ventures, bringing total funding to $94M. Here's what three years of shipping to 340 document teams taught us, and where we're pointing next.
Colophon
Marrow & Field runs Colophon's extraction pipeline across thousands of complex financial filings, helping $12B RIAs cut manual entry in half and grow without adding headcount.
Migrations are unforgiving. Here's the cutover plan we used with Pgtide and dual-write mirroring, including the two rehearsals that saved us.
Colophon
Verrata routes claims of every shape and size through Colophon, compressing a four-day audit cycle into a single afternoon.
How to build retrieval you can trust on top of Colophon's parsing and chunking primitives, so the pipeline survives its first million pages.
Unlock a decade of unstructured archives with Colophon and land it straight in your Lakewell tables.
Loomwork's builder leans on Colophon for parsing, so operations teams stand up their own review flows in an afternoon instead of a quarter.
Northrail's internal platform serves firms managing roughly $51B. Document capture, search and lineage all run through Colophon.
Pipe Colophon's document API into a Thornwood index and get usable semantic search across scanned archives in an afternoon.
Colophon
Anterra pairs Colophon's ingestion engine with nurse review, turning unstructured medical records into decision-ready summaries.
Ravelin customers have pushed more than 5.2M documents through Colophon inside their enterprise automation workflows.
Health insurance is a high-variance space. Here's how to get clean, typed fields out of scanned forms, dense tables and handwriting.
If your outputs don't match what you expected, start here. A tighter schema is usually worth more than a bigger model.
If you're weighing an in-house pipeline against a vendor, this is the honest version of the tradeoff, including where Colophon is the wrong answer.
Quarto-OCR is our open release: a smaller footprint, roughly 2.4x the throughput, and weights you can run on one GPU.
We benchmarked Quarto-OCR against Halide 2.0 and landed within 1.8 points on our internal set. Intriguing, but not yet a replacement for a purpose-built pipeline.
A behind-the-scenes look at how Colophon began powering ingestion for some of the largest companies in the world.
Table structure is where most document pipelines quietly fail. Here's what we changed in the parser.
We've published a broader open benchmark for table parsing, with 4,118 annotated pages and a public leaderboard.
Colophon now powers ingestion for several of the world's leading AI companies, and we've raised $8.6M led by Ferrous Capital to keep making human data machine-ready.
We've spent the last seven months building a document ingestion API for language-model workflows. Here's what shipped.
A look at the hidden costs of manual document work — time lost, error rates, rework — and what actually changes once extraction is automated.