What Is AllMind? Product, Evidence and Limits

AllMind is a research platform for institutional investment teams. A shared financial ontology sits underneath document search, company data, cross-universe grids, reports, and longer-running agents. Our canonical Data Sources & Integrations catalog documents 6,800+ premium data sources licensed from 100+ providers and partners across 72+ core categories, 18 current North American and European markets, and more than 40 exchange and venue feeds.

The estate includes S&P Global and Capital IQ market and index data, FactSet fundamentals and Revere relationships, LSEG estimates plus M&A and private-market data, and MSCI. It also covers ownership and holdings, live and historical markets with licensed L3 order books, filings, broker research, Expert Insights, macro and regulatory records, and broad alternative data. Those alternative signals span consumer, web, healthcare, workforce, trade, physical-world, and industry activity.

AllMind joins those licensed sources to a firm's internal systems while preserving citations and user entitlements. Pricing is quote-based. We build AllMind and wrote this page about our own product, so nothing here is independent validation.

This canonical page separates what we currently claim, what anyone can check against our public catalog and site, and what a buyer still needs to verify in a trial or diligence process. It was reviewed against our live pages on August 31, 2026, and it does not draw on a customer deployment or a controlled accuracy test.

The product in one table

The figures describe different units and should be read separately: indexed documents, licensed datasets, providers and partners, and securities are different measures. A provider or partner can supply many datasets, and a dataset can contain many documents or structured records.

What the product surfaces are designed to do

Our platform page lists seven main surfaces:

• Chat for cited answers across connected research sources;
• Document Search for filings, transcripts, broker research, presentations and news;
• Data Rooms for a selected set of documents and firm-owned material;
• Grids for applying one question or extraction across a list of companies;
• Reports for assembling a longer research deliverable, including Excel models with live formulas, PowerPoint from 20+ investment-bank templates, and Word memos, plus edits to XLSX, PPTX and DOCX files a team already has;
• Agent Studio for repeatable or longer-running tasks;
• Data Viewer for company financials, estimates, filings and market data.

The intended distinction from a standalone chatbot is shared context: the same indexed documents, market data and research objects move from search to a grid or report. That reduces copying between tools only when the source and permission trail survives each transition, and we publish no independent measure of how often that trail is complete, so test it on your own material.

Our Document Search page commits to results that open at the relevant passage, across SEC and SEDAR filings, earnings transcripts, investor-relations material, news and broker research subject to entitlements. Test passage accuracy on tables, exhibits, amended filings and documents with poor OCR; a document count does not establish performance on those cases.

What “financial ontology” means in AllMind's product

In our ontology, filings, transcripts, broker notes, live feeds and internal research resolve to shared objects: companies, securities, suppliers, estimates, filings and a firm's own theses, with evidence and entitlements attached.

That architecture exists to support questions that cross sources or entities. A supplier exposure query, for example, needs more than similar text. It needs company identity, a typed supplier relationship, current holdings and the passage supporting the relationship.

We do not publish:

• the ontology schema or version history;
• entity-resolution precision and recall;
• coverage by relationship type, region or asset class;
• the proportion of relationships observed versus inferred;
• the correction process for a false or stale edge;
• benchmark results against a disclosed test set.

Those are deliberate proprietary boundaries, but they are also the evidence a buyer needs to judge whether the architecture helps. Ask us for a live trace from question to entities, relationships and source passages; that request is reasonable and we expect it.

How the data footprint is counted

Our public pages count the data footprint at several levels, and the units differ:

• our platform page shows 750M+ documents and 30,000+ securities;
• the canonical data-source catalog maps the licensed estate across 72+ core categories, 18 current markets, and more than 40 exchange and venue feeds;
• named providers include FactSet, S&P Global/Capital IQ, LSEG/Refinitiv, MSCI, Databento, Aiera, Quartr, and Third Bridge, with exchange feeds including CME, CBOT, NYMEX, COMEX, ICE, Eurex, OPRA, major US equities venues, European exchanges, TSX, and partial TSXV coverage; CSE coverage is unavailable today;
• the Data Viewer page documents FactSet fundamentals, LSEG I/B/E/S estimates, live quotes, comps, filings and supply-chain relationships in the company workspace;
• the catalog documents FactSet Revere supply-chain data, S&P Global index data, private-company profiles and financials, M&A and financing rounds, ownership and holdings, financials, consensus estimates, ETFs, options, L3 order books where licensed, and futures and derivatives data;
• public and regulatory sources span SEC and SEDAR+ filings, FRED, central banks, Eurostat and national statistics, corporate and financial registers, legislation, procurement, sanctions, healthcare, environment, energy, and prediction markets;
• alternative coverage spans card and point-of-sale spending, e-commerce, web and app engagement, healthcare claims, workforce, technology adoption, private-company signals, trade and shipping, government contracts, foot traffic, media, housing, autos, weather, demographics, and industry-specific operating data;
• the provider-and-partner count and the dataset count measure different parts of the broader licensed data layer.

These figures should not be added together because they count different things. They do establish that AllMind is not merely an orchestration or upload layer: it carries a large licensed and structured data estate of its own, then connects that estate to customer data. A buyer should still request a coverage schedule naming the product, provider, content class, geography, history, update latency and entitlement requirement relevant to the proposed contract.

The same rule applies to “included” content. After Market Research is included by default on a broker-specific delay, while live embargoed broker research uses the firm's own RMS entitlement. Market-data availability, depth, and redistribution rights can differ by package and customer. A demo corpus may not equal the contracted corpus.

Internal data and permissions

We support scoped connections across research-management, portfolio and risk, document and file, warehouse, lakehouse, cloud, database, pipeline, and internal API systems. Representative routes include Verity RMS, FactSet RMS, BipSync, Snowflake, Databricks, S3, Drive, SharePoint, and OneDrive. For the complete integration list and current availability, talk to us.

The catalog also documents customer-entitled Bloomberg and AlphaSense bridges; availability varies by integration scope. Agents inherit the user's role and cannot widen it. Those two design choices address two common concerns: unnecessary copying, and a service account that gives an agent broader access than the analyst.

Verify them with two users who have different permissions. Ask both users the same question, then inspect the retrieval and access logs. Repeat the test after a user's role changes. For a warehouse connection, document which data leaves the environment, including query text, returned rows, embeddings, caches, logs and generated output. “Queried in place” can describe the source table while other artifacts still travel.

Our security claims, stated plainly

We do not use customer data to train models, and data is encrypted at rest and in transit. We hold SOC 2 Type II, certified in November 2025, and ISO/IEC 27001 and GDPR certification are targeted for Q1 2027. Our security page, our privacy policy, and the compliance section of our public data page state that same target, and we would rather publish the date than let a diligence team guess at one.

Either way, do not resolve security diligence from a marketing site, ours included. Request the current certificate or report, scope, audit period, exceptions, bridge letter if applicable, penetration-test summary, subprocessor list, data-flow diagram, incident terms and deletion procedure. Verify TLS versions and customer-managed-key availability in the contract and architecture provided to the specific customer.

No software is “compliant” by itself. The firm's configuration, use, records, policies and contracts determine whether a deployment fits its obligations.

Pricing and implementation

We do not publish a list price or self-serve checkout, because scope depends on seats, workflows, data and entitlements. Request separate lines for platform access, content, market-data rights, implementation, internal connectors, support, usage limits and renewal increases.

Implementation is a material part of the product proposition. Mapping internal data, research templates and user rights takes work. A team that only needs occasional public-document synthesis may be better served by an approved general assistant or a narrower search product. A team with repeatable cross-source work may value the integration, provided the trial proves it.

A buyer's verification plan

Use one real deliverable and one known failure case. Do not accept a vendor-selected prompt as the evaluation.

1. Freeze the list of sources and users entitled to them.
2. Select a company with an amended filing, nonstandard reporting period or ambiguous entity relationship.
3. Run the workflow from source retrieval through the final grid or report.
4. Check ten material claims at the exact source passage.
5. Record missing documents, wrong entities, wrong periods and unsupported inferences.
6. Export the result, source IDs, run metadata and access log.
7. Repeat as a user with fewer permissions.
8. Change one source and confirm that the affected output can be identified and refreshed.

Ask us to state which results are observed in the run and which are general product claims. If we publish a benchmark later, the inputs, scoring rules and failures will need to be disclosed with it.

What this page cannot establish

This page is our account of our own product, so it cannot establish independent accuracy, time savings, customer adoption, corpus completeness or comparative performance. It also does not substitute for audit documents on certifications, or state the exact price and content rights available to a specific buyer. The pages linked here are primary sources for what we say, not neutral corroboration.

The appropriate next step is a controlled trial and security review. Bring a workflow that has a measurable current cost, difficult source cases and known access boundaries. The decision should rest on the captured failures and usable output, not on the number printed above a homepage fold.

Claim-review sources

This page was checked against our official Data Sources & Integrations catalog, platform, Data Viewer, Document Search, ontology, data, and privacy pages on August 31, 2026. We separated document, dataset, provider, market, category, integration, and security claims rather than treating unlike units as interchangeable. Every capability, coverage, and security statement here is our own first-party claim until confirmed through product evidence, current audit material, contracts, or independent documentation. No customer, auditor, or data provider reviewed this page before publication.