Inside Kairos
Kairos joins five layers: a dated evidence store, a semantic memory, a dated graph, a team of specialist agents and the models that compute. Language models read; everything after that is checked, connected and dated. This is how it fits together, at the level we can share.
The engine
From the first cue to a forecast on the record, each stage hands the next something it can check. Solid lines run today; dashed lines are what we are building next.
01/ 07Cues
Kairos reads the biology as closely as the balance sheet: genetics and omics next to trials, regulators, filings, deals, markets and the venture record.
02/ 07Read
Extraction is the one job we give language models. Each fact keeps a fingerprint of the file it came from and the date it was published. About 700,000 so far.
03/ 07Store
A relational store keeps every fact with its provenance. A semantic memory finds evidence by meaning. Retrieval uses both, and gives primary sources the edge. A retrieval policy decides what may reach a model as evidence.
04/ 07Connect
351,316 entities and 778,170 relationships, each dated. What changed between two dates is a native question, and event sequences show which developments tend to precede which outcomes.
Dated graph · counts as of Oct 2026
05/ 07Agents
Two dozen specialist roles, from target genetics to valuation, work in parallel. Every step is reviewed, every claim is checked against the tools that agent actually ran, and disagreements stay visible.
06/ 07Reason
Models compute, from base rates that count the failures. Kairos lays out the paths and what each one rests on. You choose.
07/ 07Record
Every forecast is logged before the outcome, with a due date, and scored when it resolves.
Layers at a glance
Eight layers, each handing the next something it can check. Hover or focus a layer to see what it holds.
Solid: running todayDashed: building next
{{ lyN }} / 08
{{ lyLine }}
Next{{ lyNext }}
Cue families
Kairos reads the biology as closely as the balance sheet, and every scenario shows which threads of evidence it rests on.
Counts as of Sep 2026
43,555trials
Registries, designs, endpoints, enrolment and readouts.
FDA · EMA
Approvals, labels, designations and letters, including the Orange Book.
~700Kfacts
SEC filings, annual and investor reports, each fact traced to its file.
Licensing · M&A
Who bought, licensed or partnered what, and the patents behind it.
Patients · rivals
Epidemiology, competitors and sales.
1,128Swiss & Nordic ventures
Financing histories and outcomes, failures included.
Biology specialists
Trust by design
A retrieval policy decides what may reach a model as evidence.
Every tool answer carries a source reference, or says it is derived.
Owner-only workspaces; your data never trains a model.
Connect your own AI agents: 150+ tools.
Next Hosted beta in Switzerland, planned.
Bring a programme and the question in front of you. Kairos lays out the paths and what each one rests on. You choose.