How Modeledge works

The system at a glance

Modeledge is research infrastructure for investors who follow companies. It does three things: tracks filings, transcripts, and events on the names you cover; keeps living valuation models on those names; and saves the research you produce along the way. Everything is built around one idea — coverage that compounds. What you follow, model, and write accumulates, so next quarter starts from what you already know instead of from zero.

The system has four pillars. Each one builds on the last, and an agent connected over MCP can drive all of them.

1. Follow companies

Following is the spine of the product: your followed companies are a coverage book — the names you want to stay current on — not brokerage holdings. Every other surface keys off it.

• Your feed sorts new filings and earnings transcripts for the names you cover.
• Notifications alert you when something lands on those names.
• Valuations and refresh schedules attach to followed companies.
• An agent can answer "what's new on my names?" in one call (GetFollowedCompanyActivity).

One action makes the whole system non-empty. Follow a few companies first; everything below builds on it.

2. Investment notebooks (finmodels)

A finmodel is a structured investment notebook: assumptions separated from arithmetic. Inputs carry bounds, scenarios describe cases, formulas derive the outputs, and an evaluator computes the numbers — so a model is something you re-run, not a spreadsheet snapshot. Every model has its own web page and can be public or private.

Agents build and maintain finmodels over MCP (SaveFinModel, EvaluateFinModel), grounded in the filings and transcripts Modeledge indexes. You can turn the dials yourself with overrides without touching the math. Public models appear on the company's profile page under Models & Artifacts.

3. Valuation tracking

Binding a finmodel to a company you follow turns it into a tracked valuation: the model's headline outputs are cached on your coverage book, so the product can sort by upside, alert on changes, and keep history. A static DCF in a slide deck decays after the next print; a tracked valuation is a living position on what a company is worth, with the reasoning attached.

Agents manage tracked valuations over MCP (ManageTrackedValuation): list your coverage, check a valuation's status, bind or rebuild a model, and adjust assumptions.

4. Research artifacts

Artifacts are the memory: durable research notes and memos with their own web pages, saved from agent sessions or written directly. The point is continuity — next quarter you re-read what you concluded and why, instead of re-researching from zero. Not a chat log.

Agents save artifacts with CreateUserArtifact and search past work with SearchUserArtifacts, so earlier conclusions feed later ones.

Keeping models current

Valuations age on purpose. Refresh jobs schedule a re-underwrite when the world changes — after earnings, on a filing event, or on a calendar you set — and each scheduled date carries an honest confidence level (guess, tentative, confirmed). Every model also stores a refresh context: a runbook for how it's meant to be maintained, with methodology notes, known pitfalls, preferred sources, and a checklist.

Agents create and work these jobs over MCP (ManageValuationRefreshJob), so your models get revisited without you having to remember to.

Two ways in

The web app is free: create an account, follow companies, read filings and transcripts, and browse public models and research on company profile pages.

The Research MCP subscription connects your agent — Claude Code, Codex, Grok, or any MCP-compatible client — and the agent drives all four pillars for you: follows names, reads filings, builds and maintains models, saves artifacts, and schedules refreshes. Both sides share one account, so what the agent builds shows up in the app and what you follow in the app shapes what the agent sees.