Khatam is the on-device AEC-domain AI engine built into LocusBIM. It runs entirely on the user's machine, with no cloud round-trip and no per-inference billing. It powers cost-code matching, semantic markup search, dedup detection, and structured extraction for architecture and engineering workflows.

Is Khatam on-device or cloud-based?

Khatam is fully on-device. The model ships bundled with LocusBIM as part of normal patch updates. No project data, drawings, markups, or inference traffic leave the user's computer. This is a deliberate design choice for the Canadian data-residency (PIPEDA) requirement and extends to all customers regardless of jurisdiction.

How is Khatam different from ChatGPT or Claude?

Khatam is a small, domain-tuned model purpose-built for AEC workflows — cost-code matching, entity extraction, intent classification, semantic markup search. It is not a general-purpose chatbot and does not run in the cloud. Khatam is bundled with LocusBIM, costs nothing per inference, and works offline.

Does Khatam send my drawings to a server?

No. Khatam runs entirely on the user's machine. Drawings, markups, cost data, and inference traffic stay local. There are no API calls to a Khatam cloud service because no such service exists.

Documentation

Khatam — Overview

The AEC-domain AI engine that powers LocusBIM. On-device, no cloud, no per-inference billing. The purple khatam mark, the glow/spin interaction pattern, and what Khatam does today.

Khatam — Overview

Khatam is the AEC-domain AI engine that powers LocusBIM. It runs entirely on your machine, with no cloud, no per-inference billing, no third-party AI vendor. Khatam helps you write clearer markup descriptions, surfaces relevant cost codes, dedupes near-duplicate markups, and finds related items by meaning rather than exact wording.

Khatam is a standalone brand operated by LocusBIM Inc. LocusBIM is its flagship customer. Partner products may also embed Khatam via SDK; see locusbim.com/khatam for the product page.

Identifying Khatam in the app

Khatam is represented throughout the product by the eight-pointed purple star, the khatam mark, after the centuries-old AEC geometric pattern. Anywhere you see this mark, Khatam is either available to assist or actively working on something for you.

The mark appears as a small sparkle next to fields Khatam can help fill (markup descriptions, search bars), as a header chip in toolbars that carry AI-assisted features (the Cost Impact toolbar's confidence chip), and inside toast notifications when Khatam surfaces a suggestion you might otherwise miss.

The glow and spin pattern

Khatam uses the mark's appearance to communicate what it's doing. Learning to read this pattern at a glance is the fastest way to understand when Khatam is helping vs. when it's idle.

Static mark, no glow: Khatam is ready. Click to ask for help.
Glowing mark, no spin: Khatam has noticed something it could help with (e.g. your markup description matches a canonical AEC phrase, or a similar markup already exists). Passive prompt, nothing is happening yet. Click to see the suggestion.
Spinning mark (spin–pause–spin): Khatam is actively processing. Most inferences finish in 50–200 ms.
Purple toast with the khatam mark: when Khatam produces a result you might miss (e.g. dedup detection on save, recalculation summaries), it appears as a toast with the mark instead of a standard success/warning toast. Purple distinguishes Khatam-driven feedback from deterministic system feedback.
Copper (not purple) on a cost row: when you manually override a Khatam suggestion, the row switches from the AI-purple confidence indicator to a copper user-override colour. Copper means "you decided this, not Khatam", useful when scanning a breakdown to see what's auto and what's manual.

Tip: If you wonder whether a field has AI assistance available, look for the khatam mark. If it's not there, the feature is deterministic, and Khatam is not involved.

Khatam surfaces a range of features in LocusBIM today, from markup phrasing and search to coordination checks and cost matching. Each is documented in its own topic: markup intelligence, coordination and review, and cost intelligence.