Nomic

What is Nomic?

AI agents that run quality control and coordination review across a firm's own drawings, specifications, and project files. They read a whole set rather than one sheet at a time, which is where coordination errors live, and return a list of findings with a sheet reference for a reviewer to confirm. Covers drawing review, submittal review against the specification, RFI research over the project archive, building code checks, and plain-language search of past projects. Nomic also wrote AEC-Bench, the open benchmark for AI on coordination work, so the company has published what its own and other agents can and cannot do.

New York. Authors of AEC-Bench (March 2026), 196 tasks across nine task families on real drawing sets. Known in the ML community for the open Nomic Embed models.

Key strengths

• Works across a full drawing set and its specifications together, rather than one sheet at a time, which is where coordination errors actually appear
• every finding carries a sheet and specification reference so a reviewer can confirm it
• ready-made agents for drawing review, submittal review, RFI research, code checks, and knowledge search, so there is no prompt engineering to do
• reads drawings, specs, PDFs, markups, and BIM and IFC files without opening the authoring software
• connects to Autodesk, SharePoint, Egnyte, Box, ProjectWise, Fieldwire, and Microsoft 365
• offers on-premises or customer-cloud deployment with no data retained, per the company
• authored AEC-Bench, so its performance on this work is measurable in public rather than asserted

Common use cases

Limitations to know

Value depends on connecting your document systems first; code-check coverage and accuracy are vendor claims we have not tested; agent output still needs a reviewer