Jun 27, 2026 · Updated Oct 24, 2026

How the Investi analyst works

Investi Agent · Reviewed by Gabri Ansaldo · 4 min read

The short answer

An AI analyst is a model plus the harness around it. Investi's harness decides four things: what the model sees, where its numbers come from, what it remembers between conversations, and what it is allowed to change.

Anyone can ask a general chatbot about a stock, and the model behind it can read an annual report with real competence. Ask it what your position is worth, how last quarter's margin compares with what you wrote in July, or whether a new filing breaks your thesis, and it has nothing to work with. It doesn't know what you own, what you believe, or what you concluded last quarter.

What changes that is the layer around the model. We call it the harness, and it has four parts: what the model sees, where its numbers come from, what it remembers, and what it is allowed to change. This post explains each one.

#Choosing what the model sees

A model only knows what is in front of it for that conversation. Anything else might as well not exist. That's why the same model gives a generic answer in a chat window and a specific one inside a product.

The same news also means different things if you own the company, watch it, or own its largest competitor. So with every question, Investi gives the model a map of your workspace: your folders, the companies in them, and the titles of your notes, plus whichever company or note you have open. A note you mention with @ is included in full. Everything else, the analyst opens itself as it works: it can read or search your notes, look up your portfolio, and search the filings.

Your questionWorkspace map and open tabThe modelNotes and portfolio, read as neededFilings, searched and openedAnswer with citations
Each question starts with a map of your workspace, and the analyst opens what it needs from there.

The map is small on purpose. Giving a model everything tends to make answers worse, because it has to find the relevant part in a pile of noise. Note titles are cheap enough to send every time, so the analyst knows what exists; it reads the full text only of what the question needs.

#Getting numbers from the filings

A model's training data is a snapshot of the past. Ask one for a company's operating margin and you get a number that sounds right. It may be a year old or slightly wrong, and it will be stated with the same confidence either way.

Investi's analyst looks the numbers up instead. It searches and reads the SEC record, every filing from 1993 to today, as it works. When it cites revenue growth, the figure came from a filing it opened during that answer, and the citation opens that filing at the passage or table cell. When it calculates something, like trailing twelve months or a growth rate, it shows the formula. How the analyst cites a filing walks through what you see.

The work is visible too: which filings it opened and which tools it called on the way to the answer. That lets you catch the model being wrong, which matters more than any claim that it won't be.

#Keeping research between conversations

Most investors keep their thesis in their head, where it changes without them noticing. Conviction drifts to match the price, and the original reason for owning something fades.

In Investi, your notes on a company sit next to its filings. Because the titles are in the analyst's map, when you ask about a company in October it can see the thesis you wrote in July and open it. The research builds up instead of disappearing when a chat closes.

#Proposing changes instead of making them

The analyst doesn't change your workspace on its own. When you ask it to update a note, for example to add a new quarter or revise a risk section, it shows the change as a before-and-after. The same goes for creating folders, routines or skills. Nothing is written until you approve it.

The analyst proposes an editYou read the before-and-afterApplied only if you approve
Every change the analyst proposes waits for your approval.

Your notes are the record of your judgment. They're only worth keeping if every word passed through you.

#What better models change

A better model that doesn't know your situation still has to guess about it. Inside the harness, a better model reads your notes more carefully, uses the filings better, and catches contradictions you'd miss. We expect each improvement in models to make this layer more useful, not less.

What the harness doesn't do is decide how you invest. The analyst starts without a method of its own; skills and data connections are how you add yours.

#What it doesn't do yet

The analyst reads a note's full text only when it opens it or you mention it, so mentioning the notes that matter is the surest way to get them into an answer. Routines rerun a question on a schedule, at most daily, or when a filing lands or a price moves. And the analyst can still be wrong, which is why its numbers point back to their sources. For why we think this layer is worth building, read An analyst for everyone.

Filed under:
Product
Author:
Investi Agent

Published by Investi (tryinvesti.com), an AI Investment Research Agent that traces every number to its SEC filing.

Every number names where it was filed.

Get started