What Is Memosa?

Memosa turns a stack of deal documents — a PDF offering memorandum, an Excel underwriting model, and CoStar market exports — into a structured, fully cited institutional investment memo that you review, edit, and export from a web editor called Canvas.

It is built for commercial real-estate (CRE) investment analysts who would otherwise spend hours reading an OM, reconciling a spreadsheet, and pulling market comps by hand before the first draft of an IC memo exists. Memosa does the reading, the financial reverse-engineering, and the first draft. You do the judgment.

Note

Two inputs, one output: documents in, a cited memo out. Everything between — parsing, retrieval, multi-agent analysis, synthesis — happens automatically. You meet the work again in Canvas, where every claim carries a source marker you can trace.

Who it’s for

• Deal analysts who need a defensible first draft of an investment committee memo, fast, with the underlying evidence attached.
• Investment teams who want a consistent memo structure across every deal, with quality scored section-by-section and a readiness grade for the whole document.
• Reviewers and committee members who care less about how the draft was produced and more about whether each number traces to a source.

The two ways to start

You hand Memosa a deal in one of two places. Both run the exact same analysis pipeline underneath — the only difference is where you upload.

1. Slack — mention the bot and attach your files in an approved channel. This is the primary path for day-to-day work. See Slack intake.
2. Web Chat Intake — a guided upload flow inside Canvas, for when you’d rather not use Slack. See Web Chat intake.

Either way, the deliverable lands in Canvas as a draft memo you can open, edit, and collaborate on.

What makes it different

Memosa is not a generic “summarize my PDF” tool. Three things set it apart.

Multi-agent analysis, not a single prompt

Rather than feeding everything to one model and hoping, Memosa runs a set of parallel domain agents — financial, risk, market, comparables, exit strategy, property, construction, and more — each responsible for one part of the memo. They work under a time-budget system that guarantees each phase a share of the clock and degrades gracefully under pressure instead of timing out. The result is depth in every section, not a shallow pass over the whole deal.

Evidence citations on every claim

Every retrievable statement in the memo carries a source marker in the form [SRC:n]. Those markers resolve to the specific document the claim came from — the OM, the model, or a CoStar report. A reviewer can follow any number back to its origin. This is the core promise: the memo is traceable, not just plausible.

FormulaGraph: the spreadsheet, reverse-engineered

When you upload an Excel underwriting model, Memosa’s FormulaGraph engine resolves every cell formula into a dependency graph. It identifies the model’s top value drivers, classifies how complete the model is, and surfaces risks — circular references, hardcoded overrides, missing reserves — with cell-level provenance. Memosa reads your model the way a careful analyst would, not as a flat grid of numbers.

When sources disagree, Memosa resolves the conflict by a fixed precedence: what you tell it directly outranks the Excel model, which outranks CoStar, which outranks the PDF. (See the breakdown on supported documents.)

Where to go next

• How It Works — the end-to-end journey, from upload to approved export, at an analyst’s altitude.
• Quickstart — concrete steps to produce your first memo.

Sources

• README.md — value proposition, “How It Works” and “What’s Novel” sections, FormulaGraph and multi-agent descriptions.
• CLAUDE.md — system overview, RAG pipeline, data-source precedence (USER INPUT (100) > EXCEL (90) > COSTAR (70) > PDF (50)).
• src/utils/data_source.py — PRECEDENCE_ORDER and DATA_SOURCE_PRECEDENCE_WEIGHTS (the authoritative precedence ranking).
• src/services/intake_coordinator.py — shared Slack + web intake orchestration confirming both paths feed one pipeline.