ABOUT LOOMWELL
Existing agent work.
New learning material.
Loomwell is seeking initial partnerships with companies building finance agents to develop authorized production traces, tool calls, human corrections and supported outcomes into datasets and environments for AI labs and developers. Any supplier revenue share would depend on attributable buyer payments and a separate agreement.
For Finance Agent Companies.
The starting resource is experience already generated inside a software product. One company partnership could support recurring supply across many customer workflows, where the necessary permissions exist. Companies building platforms for clients and teams running finance agents in-house can explore earning from approved agent experience. Independent builders are welcome too; read the company partnership and payment details.
For Frontier Labs.
Our intended product combines realistic synthetic worlds, documents, tools, tasks and inspectable graders. Lab teams can discuss missing failure coverage, environment integration, permitted training and evaluation uses, and acceptance criteria.
Ten segments. One pipeline.
The product direction spans investment banking, private equity, private credit, equity research, hedge fund research, wealth management, accounting, banking ops, insurance, and trading ops & risk. We are starting with private credit and accounting, then expanding by lab demand; alpha generation and trading decisions are out of scope.
A fictional world you can inspect.
The first implemented environment is an authored synthetic data room for Northwind Logistics. An agent can read documents, query financials, perform bounded calculations and submit a structured credit memo; mechanical checks, authored-trap detection and fictional lender-policy compliance are reported separately.
Every other segment now has a structured task you can try, with grades that respond to your answer and an authored walkthrough showing the failure and correction. Explore all ten segments, plus the connected accounting pipeline.
From approved experience to licensed training products.
The proposed pilot begins with one company, an authorized source and a bounded workflow. A reviewed dataset can retain useful interaction sequences and corrections; an authorized supplier-approved pattern brief could also inform separately authored synthetic scenarios. Source access, abstraction, derivative uses, privacy and licensing require specific permission and review; finance source intake is not available through the public contact forms.
Contributors would receive an agreed share of attributable licensing revenue when approved material is licensed under their agreement. Rights, attribution, payment timing and adjustments are agreed individually; contributing does not guarantee earnings.
One pilot, then a better sourcing decision.
Match one supplier with one buyer requirement. Prepare a versioned package, ask the buyer to inspect it against agreed criteria, and record the acceptance or rejection reasons. Those reasons and the preparation effort guide the next release; a larger trace count alone is not progress.
Where the product stands.
The current examples use fictional documents and authored rules. They do not establish supplier-derived coverage, expert-calibrated policy judgment, contributor earnings or training improvement; no model is run or trained by the public demo.
We are seeking initial finance-agent company suppliers and AI lab or developer buyers. A focused pilot has to show useful additional coverage, dependable grading or preparation work it saves against a lab’s existing tasks.