Matterfact
The AI platform for investors who want an edge. Agents, data and your firm’s own knowledge, working as one research desk in your own cloud.
URL: https://www.matterfact.com/ Book a demo: https://www.matterfact.com/?utm_source=llm&utm_medium=markdown&utm_campaign=platform Last updated: 2026-10-25
Every name covered, around the clock
Matterfact is an autonomous research desk: agent swarms, frontier models, sector skills and your firm’s own data, on one platform.
- Agent swarms, the frontier of AI in finance. Fully autonomous, long-running research agents work as a team on your research goal. They build and test the investment case over many passes, the way an analyst team does, and keep going until you stop them. See https://www.matterfact.com/agent-swarms.
- 5,000+ sector skills. Specialist methods for all 163 GICS sub-industries, tuned by hand.
- The best of every frontier model. Each task goes to the model that does it best: Claude, OpenAI or Google Gemini.
- Hallucination-free answers from every call. Smart Highlight answers in management’s own words, so nothing is invented. Ask one question of every earnings call you cover and see the passages that answer it. See https://www.matterfact.com/earnings.
- Read-throughs on every name you cover. Competitor, supplier and customer news, matched to the names it moves, with how each stock took it. See https://www.matterfact.com/news.
- Your firm’s knowledge in every answer. Warehouses, licensed data, research notes and shared inboxes, connected.
- Proprietary signals from any website. From Carvana’s daily sales to every Chipotle opening, any website becomes a dataset.
- One-click playbooks. From the morning brief to a full initiation.
- 120M+ podcast episodes. What CEOs, CFOs and investors say, hours after release.
- A head start for your AI team. Single-tenant in your own cloud, with APIs and MCP to build on.
Better answers, faster, with every number checked
On Vals AI’s Finance Agent benchmark, all four frontier models scored higher in Matterfact’s agent than in Vals AI’s own harness.
| Model | Vals AI’s harness | Matterfact |
|---|---|---|
| Claude Opus 5.5 | 58.6% | 75.6% |
| GPT-6.1 Sol | 52.0% | 68.6% |
| Gemini 3.8 Flash | 61.4% | 68.5% |
| GLM 5.3 | 55.8% | 65.0% |
Per task, Claude Opus 5.5 in Matterfact was 17 points more accurate, 19× faster (33m 08s to 1m 47s) and 15× cheaper ($9.22 to $0.63). The eval report: https://www.matterfact.com/blog/towards-production-grade-evals-for-llms-in-finance
Clients
Matterfact is the firm-wide AI system at one of the world’s top four hedge funds, used on every desk. One fund used the data engineer agent to map three retail chains’ store footprints, work it had budgeted one to two months for.
They shipped in weeks what would have taken us quarters.
Head of AI, top-four hedge fund
With Matterfact, the answers are so much better, and I don’t have to check every number.
Director, Equity Research, Value Line
After an ideas dinner, I email it a thesis and it fires back the arguments that break it.
Investor, Newbrook Capital
What sets Matterfact apart is answer quality.
Vice President, Mizuho
Our vision
To make machine intelligence the edge for the world’s top investors.
Agents now do financial analysis the way a strong analyst does. They read the filings, build the model, check every read-through and work through the night. Investors keep the work machines can’t do: meeting management, running expert calls and deciding.
Matterfact is built for that division of labor: a desk of agents that knows every sector and everything your firm knows, working in swarms around the clock. Funds that work this way are better prepared, faster to act and able to see across the whole book.
Our founders come from Google DeepMind and Millennium, and our team from AlphaSense, Bloomberg, Lehman Brothers and Google. Meet the team: https://www.matterfact.com/team
What Matterfact is
Matterfact is an AI agent platform built for institutional investors: hedge funds, private equity, credit funds, venture capital, and sell-side and buy-side equity research.
Matterfact works as your team of agents specializing in different tasks. Describe the work in plain English. Research agents read filings, transcripts, and vendor data, and apply sector-specific math. Data engineering agents extract structured data from the open web and deliver it to your warehouse on a schedule.
Every number it returns carries the filing, the call, the page, or the vendor field it came from, so the work is checkable.
Who it is for
- Hedge funds, both long/short equity and multi-strategy
- Private equity and credit funds
- Venture capital
- Equity research teams, buy side and sell side
- Quantitative and systematic teams who want machine-readable alternative data rather than a chat interface
- The head of AI or head of data engineering at any of the above, who has to decide whether this sits inside their stack
How the platform is put together
Matterfact is built as five layers. You interact with the top layer; the ones underneath are what make the answers correct and governable.
Layer 1: Data
Every source your desk already pays for, plus everything Matterfact adds.
- Tier-one vendors, licensed. FactSet, S&P Capital IQ, Quartr, MT Newswire, and SEC EDGAR.
- Your own MCP server, private. If your firm exposes internal research or data over an MCP server, the agent reads it inside your boundary.
- Vendor MCP servers, managed. Matterfact manages connections to third-party MCP servers so you do not have to run them.
- Snowflake tables, governed. The agent reads and writes tables in the warehouse your firm already administers.
- External services, live. Live web sources, APIs, and feeds the agent fetches on demand.
Layer 2: Intelligence
Our fine-tuned router model matches each task to the best frontier model. We continuously evaluate its routing decisions and use the results to tune the router. You do not pick a model, and you are not locked to one vendor.
- Claude for long-horizon work: multi-step research, building a revenue bridge, running a full initiation.
- OpenAI for extraction and for writing a coherent investment memo.
- Google Gemini for long-context work and web search.
- Fine-tuned small language models that run inside your own tenant for the narrow, high-volume, repetitive classification work.
The agent also checks its own work. When a number can be verified against a second source, it verifies it.
Layer 3: Context
What the agent already knows before you ask it anything.
- Team-wide memory, shared. What your team has established about a name stays established. The agent does not start from zero on every question.
- Matterfact skills, 5,000+. The sector-specific methodologies described below.
- Your team's own skills. Your house methodology, your model conventions, and your definitions, written down as skills the agent runs the same way it runs ours.
- Ambient agents, always on. Standing agents that watch your coverage and surface a change without being asked.
Layer 4: Applications
However your desk wants to reach it.
- Chat, for interactive work.
- Email. Send a question to ask@matterfact.com and get the work back as a reply.
- Workflow builder, no code, for scheduling recurring work.
- API and MCP server. Every answer and every dataset is reachable programmatically, including from Claude, ChatGPT, and your own internal tools.
Layer 5: The platform itself
A finance-specific, vertically integrated platform rather than a general-purpose AI tool with a finance wrapper on it.
Capabilities in detail
1. Sector-specific skills (5,000+)
Our analysts and data scientists fine-tuned each skill by hand. Matterfact ships more than 5,000 of them, purpose-built for deep analysis of every kind of public company across the US, UK, and EU, and covering all 163 GICS level-4 sub-industries.
The point is that the correct analysis of a bank is not the correct analysis of a biotech or a REIT or a semiconductor company. A generic AI tool applies one framework to all of them. A skill encodes the framework the sector actually uses, including which metrics matter, how they are defined, what the traps are, and which comparisons are meaningless.
In the product, you type / in the composer and the skill library opens as a command picker. Picking a skill collapses what would be a 700-word prompt into a single command with a ticker after it. Examples:
/company-primer CMGgives an institutional-grade company overview./sotp-valuation DISruns a sum-of-the-parts valuation, segment by segment./earnings-setup NVDAlays out what the print has to clear./guidance-track-recordscores management against its own past guidance.
Your team can write its own skills, which then run the same way.
2. Research playbooks (400+)
A playbook is a one-click, multi-step research workflow. Where a skill is one methodology, a playbook is a whole piece of work, from the 6:30am brief through to a complete initiation of coverage. There are more than 400 of them, across these phases:
| Phase | What it covers |
|---|---|
| Daily briefs | Morning, weekly, and event-driven briefs, schedulable to your inbox |
| Idea generation | Quantitative screens narrowed to a workable shortlist |
| Initiating coverage | Primers, debate setup, market sizing, moat audits |
| Modeling and diligence | Revenue, margins, capital stewardship, forensics |
| Valuation | Comps, DCFs, scenarios, sum-of-parts, M&A math |
| Catalysts and events | Pre-print setup, recaps, read-throughs, calendars |
| Quality and forensics | Accounting red flags, short construction |
| Sector frameworks | Sector-specific math for when the standard playbook misses |
| Portfolio and strategy | Cross-coverage strategy work |
| Podcast insights | What experts are saying across the podcast archive |
| X insights | Sentiment, smart money, channel checks, crisis flags |
| Dashboards | Live macro, equity, and supply-chain boards |
| Web scraping | Retailer site monitors, restaurant launch trackers, data-center announcement feeds |
Playbooks can be scheduled, so the same piece of work runs every morning, every week, or on an event.
3. Agentic web scraper
Turn websites into trusted datasets with our agentic web scraper.
Custom built by former Google engineers with deep scraping experience. It handles JavaScript-rendered pages and CAPTCHA checks while honoring each website’s robots.txt crawling rules.
Most of the data that moves a thesis was never packaged as a dataset. The agentic web scraper is how you get it anyway. You point it at a market rather than at a URL, and it finds the sources, reads them, and hands back a living structured dataset that refreshes on the cadence you set.
It works on e-commerce sites, government and regulatory sites, filings, PDFs, dashboards, and job boards, in any language. Two concrete examples: track Phase III enrollment on ClinicalTrials.gov week over week, or capture every VIN listed on Carvana week over week as a used-car sales proxy.
The pipeline that runs on every source, on every refresh:
- Discover. Agents read a page the way an analyst would. They find the table, the filing, or the store locator, and map its structure.
- Extract. Pull every row from websites, PDFs, filings, and spreadsheets into the typed schema you described.
- Enrich. Geocode, translate, normalize units, and resolve entities to tickers, so the data is decision-ready rather than raw.
- Validate. Every value is grounded to the exact page, row, or cell it came from. Completeness checks run on each refresh, so you know when a source changed or went dark.
- Deliver. Land it in a dashboard, in your warehouse, or in an alert, refreshed on the cadence you set.
4. Cross-language sources
Most of the web that moves a thesis is not in English. The agent reads the source in its native language, extracts the data, and hands it back normalized, translated, and mapped to the ticker. This is coverage your English-only competitors do not have.
Japan
| Source | In English | The signal |
|---|---|---|
| 新車登録 (JADA) | New-vehicle registrations | Monthly auto demand by make, weeks ahead of OEM prints |
| リクナビ, doda | Job boards | Hiring velocity and plant requisitions as a capex nowcast |
| 食べログ, ぐるなび | Restaurant directories | Store counts, openings, and cover trend for QSR names |
| TDnet 適時開示 | Timely disclosures | Buybacks, guidance cuts, and M&A the second they post |
Greater China
| Source | In English | The signal |
|---|---|---|
| 交强险上险量 | Weekly EV insurance registrations | The cleanest read on China EV unit demand, by brand |
| 天猫, 京东 | Tmall and JD storefronts | Pricing, sales rank, and promo depth across SKUs |
| 贝壳, 链家 | Beike and Lianjia listings | Asking prices and listing price cuts by city |
| 招标, 土地出让 | Tenders and land auctions | Local-government capex and developer land banking |
Korea and Taiwan
| Source | In English | The signal |
|---|---|---|
| DART 전자공시 | Korean DART filings | Disclosures normalized, translated, and ticker-mapped |
| 쿠팡, 네이버쇼핑 | Coupang and Naver Shopping | Price and rank velocity across Korean e-commerce |
| MOPS 公開資訊 | Taiwan MOPS | Monthly revenue prints from the semiconductor supply chain |
| 104 人力銀行 | Taiwan job board | Fab hiring requisitions as a read on capex and utilization |
Continental Europe
| Source | In English | The signal |
|---|---|---|
| KBA Neuzulassungen | German auto registrations | EU vehicle demand by brand and powertrain, monthly |
| Bundesanzeiger | German filings register | Private-company financials and ownership changes |
| TED, BOAMP | EU procurement tenders | Public-sector contract flow across 24 languages |
| Doctolib | Appointment availability | Healthcare demand and clinic capacity in France |
5. What desks actually track with it
| Sector | Tracked |
|---|---|
| Restaurants and QSR | Store locators and unit counts; new openings and closures; reservation availability (OpenTable); menu pricing deltas |
| Consumer and retail | Shelf pricing across UPCs; out-of-stock rates; DTC storefront launches; marketplace rank velocity |
| Pharma and health | ClinicalTrials.gov enrollment; FDA 483s and PDUFA dates; the FDA drug-shortage list; Orange Book patent expiries |
| Labor and industrial | Job postings and hiring velocity; fab and plant requisitions; dealer inventory and days-on-lot; permit and filing trackers |
| Energy and utilities | ISO prices and demand (ERCOT, PJM); generation mix shifts; interconnection queues; outage and capacity maps |
| Filings and regulatory | Global filings, normalized; Federal Register actions; NHTSA recalls and EPA enforcement; 8-Ks and IR-page changes |
6. Podcast intelligence
Matterfact indexes the complete full-text transcripts of more than 120 million podcast episodes across roughly 700,000 shows, transcribed, tagged, and joined to tickers. This is proprietary data that no terminal carries.
Podcasts are where CEOs, CIOs, operators, and fund managers speak at length and with far less compliance filtering than on an earnings call. You can search any company, executive, ticker, or theme across the whole archive and get sourced, quotable answers, or set a standing alert for the moment someone mentions a name you cover.
The same data is available as machine-readable annotations for quantitative teams: one JSON record per episode and subject company, with speakers typed by role and passages scored for stance and hedging. See https://www.matterfact.com/annotations.
There is also a podcast MCP server, so you can query the archive directly from Claude or ChatGPT. See https://www.matterfact.com/blog/connect-matterfact-to-claude.
7. Dashboards on prompt
Describe the analysis in plain English and the dashboard renders. Margin bridges, scenario decks, catalyst calendars, peer comps, and location heatmaps, all built from the underlying data rather than pasted in, and all refreshing as the data does.
8. Sources you already trust
Every figure carries the filing, the call, or the vendor field it came from. The licensed sources behind the platform include FactSet, S&P Capital IQ, Quartr, MT Newswire, and SEC EDGAR. Coverage of venues spans the US, UK, and EU.
9. Built to accelerate your AI team
Deployed at some of the world’s largest hedge funds, Matterfact helps internal AI teams move faster. Extend your platform with financial skills, connected data, and research workflows.
Customer perspective
Matterfact has been an exceptional partner in accelerating our AI efforts. The platform is so extensible and fits our needs so closely, it feels like our own internal AI platform.
Managing Director, AI Technology, $20B+ hedge fund
Built for your fund. With or without an AI team.
Use Matterfact as your fund’s AI platform, or connect it to the systems your AI team is already building. Your research, data, and workflows stay private to your firm.
The Snowflake agent
Connect the warehouse your firm already governs. The agent reads your internal tables alongside vendor data, and writes every dataset it builds back as governed tables under the roles and grants you already administer. There is no parallel permission system and no export step. Schema migrations and refresh schedules are managed for you, and the resulting tables query like any other table in your warehouse. Typical pattern:
| Table | Mode | Cadence |
|---|---|---|
alt_data.mcd_store_counts |
Writes | Weekly |
alt_data.de_auto_registrations |
Writes | Monthly |
research.coverage_kpis |
Reads | Live |
vendor.consensus_estimates |
Reads | Live |
Datasets can also land in S3, BigQuery, or Databricks.
Security
Private to your fund, with the controls your CISO expects. Matterfact is single-tenant, SOC 2 Type II audited, penetration tested and vetted by CISOs at top-tier hedge funds:
- Single-tenant: one isolated deployment per fund, in our cloud or yours
- SOC 2 Type II, independently audited, with third-party penetration tests
- Zero data retention: model providers keep nothing, and your data trains no one else’s model
- Role-based access through your identity provider, down to the file
- An audit trail of every query, source and output, with a citation for every claim, exportable to your SIEM
- TLS 1.3 in transit, encrypted at rest
We share the report and support DDQs and vendor security reviews under NDA.
Deployment
Matterfact runs single tenant. One isolated deployment per firm, never a shared workspace. It runs on AWS or Google Cloud, or inside your own VPC or on-premise if your requirements are stricter, with default-deny egress and a full audit trail of every source the agent touched. Your data never trains anyone else's model. Full detail is at https://www.matterfact.com/security.
Frequently asked questions
Is Matterfact a chatbot? No. You can talk to it, but the output is working analysis: models, datasets, dashboards, and memos, each grounded to a source you can open.
Which AI model does it use? Whichever one handles the task best. Tasks are auto-routed across Claude, OpenAI, and Google Gemini, plus fine-tuned small models running inside your own tenant. You are not locked to a single model vendor.
Does it replace my data vendors? No. It connects to them. FactSet, S&P Capital IQ, Quartr, MT Newswire, and SEC EDGAR are read alongside your internal warehouse tables and anything the web scraper collects.
Is my data used to train models? No. Each firm runs in its own isolated deployment, and your data never trains anyone else's model.
Can it run inside our own cloud? Yes. AWS or Google Cloud, or your own VPC or on-premise.
How is Matterfact priced? Matterfact is an enterprise product, licensed per firm on an annual contract. There is no free tier and no self-serve signup. Pricing depends on the size of the team, the data sources you connect, and which deployment you choose, so it is scoped on a call. Book a demo to get a quote.
What is the difference between a skill and a playbook? A skill is one methodology, invoked as a slash command. A playbook is a multi-step workflow that chains many of them into a whole piece of work, and it can be scheduled.
Can we write our own skills? Yes. Your house methodology, definitions, and model conventions can be written as skills, and the agent runs them exactly the way it runs Matterfact's own.
How do I get data out? Snowflake, S3, BigQuery, or Databricks, plus a REST API and an MCP server. Dashboards and email alerts are also first-class outputs.
Related pages
- Home: https://www.matterfact.com/
- Podcast intelligence: https://www.matterfact.com/podcasts
- Podcast data for quants: https://www.matterfact.com/annotations
- Security and single-tenant deployment: https://www.matterfact.com/security
- Blog: https://www.matterfact.com/blog
- Podcast newsletters: https://www.matterfact.com/newsletter
- Site index for LLMs: https://www.matterfact.com/llms.txt
Getting started
Tell us a company you cover and the sources you wish you were tracking. We will run the playbook, build the dataset, and show you the first alert, with every figure sourced.
Book a demo: https://www.matterfact.com/?utm_source=llm&utm_medium=markdown&utm_campaign=platform&utm_content=demo