The loan runs itself. You run the decisions.
Qualr AOS is an agentic origination system for residential and commercial lending. Agents carry the file from first touch to post-close and beyond — assembling, extracting, reconciling, and chasing — and stop at every gate where a person should decide.
- 8 stages
- Intake through post-close retention
- Human gates
- On every decision that carries risk
- Fully cited
- Every value links to its source page
- Your cloud
- Google Cloud or AWS, your account
A rules engine follows a flowchart. An agent has a goal.
The difference shows up in the exception rate. A flowchart escalates everything unusual to a human. An agent re-plans, and escalates only what actually needs judgement.
Goal-driven, not step-driven
An agent is given the outcome — "this file is complete and underwritable" — and works out the steps. A rules engine breaks the moment a file does not match the flowchart; an agent re-plans.
Every action is cited
Each extracted value links to its source page, and each guideline call links to the section it came from. A reviewer verifies in one click instead of reopening the PDF.
Gates are structural
Credit, compliance, and anything that touches a borrower stop for a person. That is architecture, not a toggle somebody can switch off to hit a turn-time target.
Fully reconstructable
Every prompt, retrieval, tool call, and response is logged. When an examiner asks why a file went the way it did, the answer is a record rather than a recollection.
Reversible by design
Agent actions are proposals against the loan record until a human commits them. Nothing an agent does is unrecoverable.
Measured, not assumed
Every stage reports how long it took and how often it needed a human. You can see exactly where the agent is earning its keep and where it is not.
Eight stages. The agent works all of them.
Most systems stop at funding — which is exactly where the next loan gets created. Each stage below states what runs on its own and what still stops for a person.
- 01
Intake
Agent does
A visitor is captured, scored against your criteria, and routed to an owner in under 90 seconds. The application then drafts itself from the documents the borrower already has — bank statements, entity paperwork, income records — so the borrower confirms facts instead of typing them.
Human decides
Who owns the relationship, and whether the scenario is one you want.
LeadGrab LiteAPP - 02
Assemble
Agent does
Every document is classified the moment it lands and filed against the condition it satisfies. The agent works out what is still missing and requests it from the borrower or broker directly, then keeps asking until the file is complete.
Human decides
Nothing routine. Exceptions and unreadable documents are escalated, not guessed.
Diver GSD
- 03
Extract & reconcile
Agent does
Rent rolls, operating statements, and bank statements become structured data. The agent cross-checks the same fact across every document in the file and surfaces the disagreements — the discrepancies that normally surface late, at underwriting, with a rate lock already burning.
Human decides
Reviewing flagged conflicts. Every value links back to the page it came from.
Diver Bespoke SLMs - 04
Apply guidelines
Agent does
Your credit policy is applied to the assembled file. The agent drafts the condition set, cites the guideline section behind every one, and writes up the exceptions it thinks are worth asking for.
Human decides
The credit decision. Always. The agent drafts and cites; it does not approve, deny, or grant an exception.
Qualr AOS Bespoke SLMs
- 05
Price & lock
Agent does
Scenarios run across your investor shelf at once, including the non-QM programs conventional engines handle badly. The agent normalises the results into one comparison and prepares the lock request.
Human decides
Which product, which price, and when to lock.
Qualr AOS LendLAB - 06
Clear conditions
Agent does
Conditions are chased on a schedule, cleared automatically when a document satisfies them, and re-run against the file when anything changes. Status stays visible to the broker without an email thread.
Human decides
Signing off conditions that require judgement rather than a matching document.
GSD Diver
- 08
Post-close & retain
Agent does
Trailing documents are collected, QC findings are routed, and the borrower moves into portfolio monitoring. The agent watches for the events that create the next loan — maturity dates, rate movement, portfolio milestones — and raises them before a competitor does.
Human decides
What to do about the opportunity once it is on the desk.
RainMaker GSD
Layer it on, or let us build the whole thing.
Stack on top of your LOS
Keep your system of record. AOS runs the agent layer over MeridianLink, LendingPad, or Encompass, and connects to anything without a modern API as a custom build. Nothing gets ripped out, and you can prove the workflow on one channel before it touches the rest.
See what it connects toOr we build you a custom end to end
If the system of record is the actual problem, we replace it. Origination, documents, pricing, conditions, and delivery on one loan record, built to your credit policy rather than to somebody else's product roadmap — and deployed into your own Google Cloud project or AWS account.
Talk about a custom buildQualr AOS questions
What is an automated origination system (AOS)?
An AOS is a loan origination system with an autonomous agent layer on top of it. A traditional LOS is a system of record with a workflow attached — it holds the file and tracks what stage it is in, but a person still drives every step. An AOS is given the outcome instead, and works the file toward it: assembling documents, extracting and reconciling data, applying guidelines, chasing conditions, and delivering the closed loan, stopping at defined gates for the decisions that require a human.
How is this different from workflow automation?
Workflow automation executes a flowchart somebody drew in advance, and it breaks the moment a file does not match it. An agent is given a goal and re-plans when reality differs from the plan — a missing document, a discrepancy between a lease and a bank statement, an entity structure nobody anticipated. In practice the difference shows up in the exception rate: the flowchart escalates everything unusual to a human, the agent escalates only what actually needs judgement.
Does the agent make credit decisions?
No, and it is built so that it cannot. The agent assembles the file, applies your guidelines, drafts the condition set, and cites the guideline section behind each one. The credit decision, the exception, and the authorisation to fund are human gates enforced in the architecture rather than by policy. An agent that could approve a loan would be a compliance problem, not a feature.
What does "cradle to grave" actually cover?
Eight stages: intake, document assembly, extraction and reconciliation, guideline application, pricing and lock, condition clearing, close and delivery, and post-close retention. The last one matters more than it sounds — most systems stop at funding, so the events that generate the next loan (maturity dates, rate movement, portfolio milestones) go unwatched.
Can we run AOS on top of our existing LOS?
Yes, and that is the usual starting point. AOS reads and writes to MeridianLink, LendingPad, and Encompass, so it can run the agentic layer over the system of record you already have. If the system of record is itself the problem, AOS can replace it — that is a custom end-to-end build rather than a layer.
Where does it run?
On Google Cloud or AWS, and it can be deployed into your own project or account rather than a shared tenancy. For regulated deployments your data and your models stay inside your region and your compliance boundary, and audit logs deliver into the tooling your security team already uses.