Inside FinCatch Studio — the Operating Environment, in Plain Words | FinCatch Blog

Published 9 September 2026 · Read time ~5 minutes · For the analyst who'll use Studio every day, not the buyer who's evaluating it.

Studio is officially out of beta. Since May 2026 we ran Studio as a private beta — internal testing, a handful of early users, and a fast feedback loop that surfaced the gaps between v0.1 and something worth shipping. We changed the architecture in late 2025, redesigned the market graph through 2026, and rebuilt the product around the cross-check at every step. Today FinCatch Studio's first version is generally available — call it v1.0, with the architecture (Financial Minds), the operating environment (the workspace + surfaces), and the cross-check (the mechanism that makes every answer citable). The market graph covers ~1,500 companies; the team's prior views and procedures are wherever the team has captured them; Agent Memory is wherever you write for the agent.

This post is a walkthrough — for the analyst who'll use Studio every day and wants to know what the discipline actually does. The architecture argument is on the company site; the use cases are on the Use Cases page. This is the operational view: what it's like to open Studio on a Tuesday, type a question, and get an answer you can put in front of the IC.

The cycle — every question goes through it

Every time you ask Studio something — "compare the margins of these three companies over the last eight quarters", "what's the latest view our team holds on this name", "draft an earnings reaction for the report we just got" — Studio runs the same five steps. Here's what happens:

Plan. Studio reads your question and decides what counts as a good answer. "Compare margins" needs company entities and financial data. "Latest view" needs the team's stored views. "Draft an earnings reaction" needs the press release plus the team's prior views and earnings-reaction procedure.
Pull. Studio pulls what it needs in parallel — the market graph yields the relationships, the personal memory graph yields the team's views, the library yields the consensus documents, Agent Memory yields your hand-curated notes, Skills yield the procedure to follow.
Cross-check. At every step — every entity, every event, every relationship — Studio compares what each source says. Where they agree, the answer is consensus. Where they disagree, Studio surfaces the disagreement as the gap you should look at.
Draft. Studio composes the answer (the memo, the comparison table, the chart caption). Every figure carries the source citation. Every assumption is editable.
Hand off. The draft lands in the workspace. You review, comment, edit. The next cycle inherits the prior output — not as a saved file but as the active context.

It's fast — single-digit seconds for most questions. The discipline is in steps 2 and 3: the cross-check at every step is what distinguishes Studio's answer from a chatbot's.

The cross-check — the disciplining bit

Most AI tools for finance retrieve a corpus and summarise. Studio does something different: at every step of the agent's walk through the market graph, it compares what each of its sources say.

A concrete example. You ask Studio: "What does the latest earnings release imply for our covered semiconductor names?" Studio walks the market graph from the issuer (the foundry) to its known customers (the names you cover). At each step — foundry → customer → catalyst → guidance → exposure — Studio compares:

Your team's prior views — does the team have a view on this entity or event? If yes, that view is in the answer. "The team thinks the foundry's capex reduction is overdone; the customer should benefit."
Consensus documents — does the broker library say anything about this entity or relationship? If yes, the broker note or research summary is cited. "Goldman published a note on the foundry's pricing power last month."
The market graph itself — is this relationship a confirmed link in the graph? If yes, the propagation is structural; if not, Studio flags the relationship as inferred and the answer is qualified.
Agent Memory — do you have a hand-curated note about this? If yes, your view is in the answer.
The relevant procedure (Skill) — which workflow applies here? The earnings-reaction Skill, the supply-chain Skill, the morning-brief Skill — each one defines how Studio composes the answer for that kind of question.

Where all five sources agree, the answer is consensus. Where they disagree, Studio surfaces the disagreement as the gap you should look at — "the broker library says the foundry's pricing holds; the team's prior view says pricing breaks; Agent Memory says you've been watching this; here's the evidence for each side." The cross-check doesn't resolve disagreement. It surfaces disagreement so you can judge.

A chatbot with a search box summarises the consensus. Studio runs the cross-check at every step and surfaces what the consensus misses.

Where Studio gets its five sources

Studio doesn't ask the model to remember anything. It pulls from five places — every time:

The market graph. When you ask about a supply chain, a propagation, or a second-order effect, Studio walks the graph and shows you the relationships. You don't see the graph; you see the walk. The walk produces the propagation map, the catalyst chain, the relationship between two entities you mentioned in passing.
The team's prior views. The team's thesis on Company X, the analyst's working view on Company Y, the conviction history that says "we've been wrong on this before". Studio pulls these from the team's memory and puts them in the answer — not as an opinion to push, but as the position the cross-check challenges.
Consensus documents (the Library). Broker reports, research summaries, IR releases, anything you've filed or Studio has indexed. Studio pulls from the Library when it needs the consensus position on an entity or event.
Agent Memory. Your hand-curated knowledge — different from the team's prior views, which Studio writes from your actions; Agent Memory is you writing for the agent. A thesis file, a mental-model file, a divergence file. Studio reads Agent Memory before it composes the answer.
The relevant Skill. Earnings-reaction Skill, supply-chain Skill, covenant-review Skill, scenario-analysis Skill. Each Skill defines how Studio composes the answer for that kind of question. You can author your own Skills from notes or a book — Studio learns the procedure and applies it from the next cycle onwards.

Studio pulls from these five sources in parallel, then cross-checks at every step. The five coexist; they don't stack.

Where the work lands — six surfaces

The cycle produces an answer; the surfaces are where the answer lands. Six surfaces ship at launch; more follow as Studio grows:

Live spreadsheet. Every calculation is a formula, every edit propagates, every assumption is editable. Bull/bear scenarios, trend extractions, covenant register, the analyst's ad-hoc models. Auditable because every formula is visible; editable because you change assumptions directly; reproducible because the workspace loop carries the formulas forward into the next cycle.
Slide deck editor. IC-ready decks from any output. The slide deck imports the same anchored data as the report — every chart, every number, every citation carries the same source.
Email agent. Native Gmail and Outlook ingest; emails become citable Library objects. Filing an IR release by forwarding it to Studio is the same as filing it manually.
Chart builder. Price charts, time-series, peer-comparison visuals. Charts are drawn from the same canonical sources as the report; the chart and the citation are consistent.
Report editor (long-form). Earnings reactions, quarterly reviews, investment memos, due-diligence notes. Studio drafts from the same five-source cross-check; every figure, every citation, every claim is anchored to the source. You review the draft, mark what to change, and Studio re-drafts.
Connectors. Telegram, Gmail, Google Drive. Every connector lands material in the Library as a citable object. The connector is the entry point; the architecture is what happens after.

Skills — the procedures

Skills are how Studio knows what kind of question you've asked. A Skill called "earnings reaction" walks Studio through the workflow: pull the press release, standardise the figures, query the team's prior views, draft the one-page memo, write the verdict per view into the workspace. A Skill called "supply-chain second-order" is different — propagation, exposure, falsifiers. A Skill called "covenant review" is different again — breach detection, source attribution.

Studio ships with pre-built Skills for the standard finance workflows. You can author your own Skills from notes, a book, or a procedure you've been running manually — Studio learns the procedure and applies it from the next cycle onwards.

Three jobs Studio is built for

Studio's architecture and its surfaces are designed for a small set of recurring jobs that take up most of the analyst's day. Three illustrative jobs:

An earnings reaction in 30 minutes, with every figure defensible to a compliance reviewer. Studio ingests the press release and the transcript, pulls the team's prior views on the name, and drafts a one-page memo — every figure anchored to the source, every citation resolvable to a passage.
Tracking the second-order effects of a supply-chain shock. A foundry announces a capacity constraint; Studio walks the market graph from the foundry to its known customers to its customers' customers to the catalysts each is exposed to. The propagation map surfaces in the workspace; the team gets a flagged gap where the consensus library has nothing on the relationship.
Re-testing your team's prior views when a new event lands. A catalyst, a guidance change, or a data print hits a covered name. Studio matches the new event against the team's stored views; the agent writes a short verdict per view (strengthens, weakens, invalidates, or unchanged) into the workspace. The next morning's brief already includes the change — no manual re-reading.

See all six launch use cases (insight discovery / knowledge management / workflow automation) on the Use Cases page.

Where the discipline lives

Studio is the architecture and the surfaces — the workspace, the live spreadsheet, the slide deck, the email agent, the chart builder, the report editor. Studio does the cycle — every cycle starts where the last ended, the next cycle inherits the prior output, the team's accumulated work is the substrate. The cross-check at every step is what makes the answer citeable, reproducible, and re-testable. Studio cannot replace judgment; the opinion is the analyst's.

For what Studio is not (chat wrapper, single-LLM tool, search tool, trading tool, Bloomberg replacement, judgment replacer), see the FAQ on the Use Cases page. For the production-deployment discipline (data sources, Skills as user procedures, LLM limitations, licensed-professional context), see the Terms of Using Studio on the Studio site. For the Studio-specific data-handling terms, see the Studio Privacy Notice. For the company-site data-handling terms (cookies, marketing, support), see the company-wide Privacy Policy.

Where to start

If you're new to Studio, two entry points cover most of the value:

Agent Memory. Write one thesis file — what you currently believe about a covered name. Drop it in. Studio's next cycle will read it as the position the cross-check challenges. Over time, your Agent Memory becomes the team's mental model — the layer no chatbot with a search box has.
A Skill that mirrors a procedure you already do. If you run an earnings reaction, write a Skill that walks Studio through your process. If you write a weekly thesis health-check, write that Skill. The first Skill takes an hour; the second takes ten minutes; the third is muscle memory.

Studio is at v0.1. The market graph covers ~1,500 companies; the team's prior views are wherever the team has captured them; the Library is whatever the team has filed. The architecture will grow as you grow it. The cross-check is the constant — every cycle, Studio pulls from all five sources, compares what each says, and the answer is either consensus (where they agree) or a named disagreement (where they don't). That is the discipline. That is what Studio is built around.

Welcome.

— Lok Kan Chan (CEO) and the FinCatch team