── ── Data & methodology
How is deciqAI's knowledge base built?
deciqAI's answers draw on 26,724 companies profiled and 1.3M+ raw data entries. This page defines what each number counts, where the data comes from, how often it updates — and how to verify it yourself.
The numbers, defined
Every figure we publish is a count from a live system, not a marketing estimate. Here is what each one means — and how to check it.
Company trajectories — founding to outcome, the wins and the losses — profiled in our knowledge base. Over 8,500 of them carry deep annotation: 41,000+ growth-phase records and 107,000+ logged strategic actions, coded across 8 company phases and 12 operating dimensions.
It's a count, not an estimate — the number of profiles at the last import batch.
The conservative floor. Our retrieval index and structured databases sum to over 1.5 million entries; we publish 1.3M+ and round down. The full breakdown is below.
See “What's inside the 1.3M+” for the composition.
The thinking frameworks every agent is wired with. All 163 are open source under MIT — you can read exactly what your agents reason with, skill by skill.
Inspect them at github.com/deciqAI/knowledge-skills.
The full operator roster — all live, no waitlist. Each agent's job, inputs, and deliverables are listed publicly.
Browse the roster on the Features page.
── ── Composition
What's inside the 1.3M+
Vectorized retrieval index
700,000+ entries
Knowledge chunks embedded at 1,024 dimensions and retrieved on every answer. Includes company intelligence distilled from the 26,724 profiles, global market knowledge across 26 countries, VC ecosystem profiles (firms, accelerators, operators) across 25 countries, industry & location knowledge, and 271 public research reports (UNESCO, OECD, and others) distilled into 3,400+ chunks.
Decision-outcome records
830,000+ records
Real education and career decision outcomes — applications, admissions, and paths taken — across 15,000+ institutions in 30 countries. These power the education and life-path products and ground the platform's outcome modeling.
Summed, these systems hold over 1.5 million entries. We publish 1.3M+ and round down — the headline should always be the number we're most sure of.
── ── Sourcing
Where the data comes from
Public & open data
Government business registries, public company filings (including SEC Form D), and open data from international organizations such as UNESCO and the OECD.
In-house annotation
Analysts profile company trajectories into phases, dimensions, and strategic actions through a review pipeline with quality gates — this is the labor behind the deep-annotated subset.
Licensed datasets
Where public data ends, we license commercial datasets. Some licenses don't permit public attribution — those sources we describe by category only, and never claim as our own research.
── ── Freshness
How often it updates
News and filings feeds are ingested into the knowledge base every day.
An automated knowledge-gap analysis flags thin coverage and queues the next import batch.
Company-annotation batches are imported and vectorized as they clear the quality gate — the headline counts move with each batch.
── ── Your data
How your data is treated
The knowledge base above is what deciqAI knows. Your data is a separate matter — here is exactly where the line sits.
No training. Ever.
deciqAI never trains or fine-tunes AI models on your data. Models are used for inference only — your inputs produce your outputs, and nothing else.
Your workspace stays yours
The shared knowledge base is built from public data, licensed datasets, and our own annotation work — never from customer workspaces. Nothing you upload, connect, or discuss enters it.
You press the button
Outbound actions — sending an email, creating a document, touching a calendar — queue for your explicit approval. The agent proposes; you decide. And you never have to connect a system you're not comfortable with.
── ── The fine print, up front
What these numbers are not
Not customer counts
The 26,724 companies are research subjects in the knowledge base, not deciqAI users.
Not endorsements
A company being profiled means we studied its trajectory — including the failures. It implies no relationship with deciqAI.
Not a guarantee
The data grounds the analysis; it doesn't replace judgment. deciqAI shows what each path costs — you still make the call.
Frequently asked questions
What does “26,724 companies profiled” mean?
It's the count of company trajectories — founding to outcome, wins and losses — in deciqAI's knowledge base at the last import batch. Over 8,500 carry deep annotation: 41,000+ growth-phase records and 107,000+ logged strategic actions across 8 company phases and 12 operating dimensions.
What does “1.3M+ raw data entries” actually count?
Two systems: a vectorized retrieval index of 700,000+ knowledge entries (company intelligence, market research across 26 countries, VC ecosystem profiles across 25 countries, distilled public reports) and 830,000+ real decision-outcome records across 15,000+ institutions in 30 countries. Summed they exceed 1.5 million; deciqAI publishes 1.3M+ as the conservative floor.
Why is the number so precise — why 26,724 and not “25,000+”?
Because it's a count, not an estimate. The knowledge base is imported in audited batches, so the profile count is exact at any point in time. It moves with each batch, and this page is updated when it does.
Can I verify any of this myself?
Yes. The 163 skills every agent is wired with are open source under MIT at github.com/deciqAI/knowledge-skills — read them skill by skill. The full 58-agent roster is public on the Features page. And this page defines exactly what each headline number counts.
How often is the knowledge base updated?
Daily for news and filings feeds, weekly for automated knowledge-gap analysis, and rolling imports for company-annotation batches as they clear the quality gate.
Does deciqAI train AI models on my data?
No. deciqAI never trains or fine-tunes models on customer data — models are used for inference only. Your workspace data is also never added to the shared knowledge base, which is built exclusively from public data, licensed datasets, and deciqAI's own annotation work.
Do agents act without my approval?
No. Outbound actions — sending an email, creating a document, changing a calendar — are queued as pending until you explicitly approve them. You can reject any proposed action, and you never have to connect a system you're not comfortable sharing.
Start free. Pay when it pays off.
Spin up your Operator in under 10 minutes. No card required to start.
Start free