The questions lenders ask first.

No. Apsis Line is a software platform, a mortgage origination point of sale and workspace. We don’t originate, broker or fund loans, and we don’t make credit decisions. Your institution remains the lender, and your loan origination system (LOS) remains the system of record.

No, it’s the opposite. We built Apsis Line as an intelligent middleware layer above the LOS you already run. It pushes complete Mortgage Industry Standards Maintenance Organization (MISMO) 3.4 applications into the LOS, and we’ve built it to order vendor services through the LOS’s service channels and generate disclosures through it. It builds on the investment you’ve already made in the LOS, and we don’t duplicate it.

Yes. The platform is LOS-agnostic by architecture. It speaks MISMO 3.4, the industry-standard interchange every major LOS accepts, and every vendor connection is an adapter seam. Our integration is deepest today on Encompass from Intercontinental Exchange (ICE). Loan create and field-level read/write are live code, and we’ve built Partner Connect service orders and disclosure generation to ICE’s published contracts. They run simulated until you connect your sandbox, and none of it has run against a live Encompass instance yet. We’d wire other LOS platforms through the same seams. If you run something else, that’s an integration conversation, and it doesn’t disqualify you.

In the loan officer workspace, disclosures hard-gate on four service orders: soft credit, hard credit, flood determination and an automated compliance review. The review has to affirmatively clear, so a failed, pending or errored review holds the file. Closing costs gate the same file a second way, through escrow readiness: the Loan Estimate’s escrow figures come from an ordered fee schedule and fail closed without one. On the borrower self-serve journeys, cash-out and purchase alike, flood, fees, escrow readiness and the compliance review all have to clear before the platform orders any disclosure. We hold a purchase with no property address yet until it has one.

Yes, on optional surfaces, and none of them is in the regulated path: a guideline assistant for your loan officers, a borrower question panel, a staff co-pilot, and document intelligence that classifies an uploaded document and extracts its fields. Each surface stays off until you configure it, and enabling one doesn’t enable any other. On the question-answering surfaces, retrieval runs first over a bounded corpus, we ask the model to phrase and cite the retrieved passages, never to reason, and our code validates every draft answer after the model returns or a scripted refusal ships instead. Document intelligence picks a document type from a closed list, our code re-checks it, and if anything fails it doesn’t guess: it falls back to a labeled simulation.

No. No AI or machine learning (ML) model produces, influences, or ranks any output that determines a regulated outcome: credit decisioning, pricing, valuation, adverse action, disclosure content or timing, or a servicing action. Continuous integration enforces that, so we don’t have to assert it: a gate walks the transitive import graph out from every module that owns or feeds a regulated outcome and fails the build if a model endpoint becomes reachable from any of them. It’s a static import reachability property, so it doesn’t cover a dynamic import, a runtime-assembled Uniform Resource Locator (URL), or a model reached through stored data. The gate prints that limitation on every run and our disclosure states it.

Fannie Mae Lender Letter (LL) 2026-04 requires you to manage risks and governance of subcontractor and vendor use of AI/ML no less protectively than your own, and both agencies can require you to disclose the types of AI/ML in use, the purpose and manner of that use, and the safeguards around it. We add our model surfaces to your AI inventory (we don’t claim to shrink it) and hand you each one already tiered and scoped, with a written governance disclosure and named subprocessors, so you can discharge the obligation. Read this part carefully: Freddie Mac Guide §1302.8 contains no vendor clause itself. That reach comes from the scope of the bulletin that introduced it.

No. We don’t train or fine-tune any model on your data, your borrowers’ data, or ours. The governance disclosure names our runtime model provider and exactly what each surface transmits to it. For the staff co-pilot’s answer relay, that’s staff-written text carrying no case facts at all. For document intelligence, it’s the uploaded document itself.

A self-guided journey that feels like a conversation with a seasoned loan officer: their real numbers (live home-value estimate, soft-credit balances with consent), an honest side-by-side of their options, math they can open and check, and e-sign disclosures at the end. A borrower who wants a human at any point gets one, with the file intact.

Yes. Tenancy is configuration work, and it doesn’t take code. Your brand, palette, program overlays and vendor connections resolve from a registry entry, and borrower traffic runs under your domain. We deliberately keep the disclosure and consent surfaces out of theming, so they look the same for every tenant.

Server-side only, in managed environment configuration and vaults. No application programming interface (API) key or Open Authorization (OAuth) secret ever ships to the browser. We label every vendor result with its source, and the server, never the client, decides whether a call runs simulated or live.

It shows up in the loan officer workspace pipeline with everything intact: entries, consents and service results. The loan officer (LO) continues the file, and nothing restarts. A stalled journey can also enter an automatic follow-up cadence that carries the context the borrower left behind.

A first walkthrough is about thirty minutes: the borrower journey end-to-end, the workspace, the compliance gate refusing an unclean file, and the exact MISMO 3.4 file we push to your LOS (simulated until you’ve connected your sandbox). We run deeper integration evaluation, covering field mappings, service orders and security review, under a nondisclosure agreement (NDA) with your team.