The intelligence
layer for surgery.
Trigone builds AI that perceives, understands, and reasons about the operating room. Labelling objects is not understanding — meaning depends on the procedure, the phase, and what came before. Every conclusion carries the evidence behind it, and the surgeon keeps authority over all of it.
In development with clinical rigor. Not a cleared medical device.
The operating room produces medicine’s richest data.
Almost none of it is understood.
Surgical AI today mostly means isolated computer vision: models that draw boxes, label pixels, or flag single events. Each sees a fragment. None of them understands the operation.
Detects objects — in isolation
Segments pixels — without meaning
Flags events — without context
A detection without context is a data point.
Surgery demands understanding.
The same instrument, the same anatomy, the same movement can be routine in one moment and safety-critical in the next. What separates them is everything around them — procedure, phase, history, evidence. That context is exactly what isolated models cannot see.
From seeing
to understanding.
Trigone connects what isolated models keep apart. Detections become entities. Entities form relationships. Relationships accumulate into an interpretation of the procedure itself — one that carries its evidence with it.
- Seeingan instrument, a structure, an event — separately
- Understandingwhat this moment means, in this procedure, right now
A system that reasons,
not a model that labels.
Trigone is built as layered intelligence. Each layer consumes the one below it and is inspectable on its own — so the system’s conclusions can always be taken apart and checked.
Select a layer to inspect it. Conceptual architecture — the interesting parts live below this level of detail.
What it takes to understand
an operation.
Surgical understanding is not one capability. It is many, working together — each one grounded in what the system can actually observe.
Anatomy
Which structures are in view, how confidently they are identified, and how that identification evolves.
Instruments
Which tools are present and active, what they are doing, and what they are doing it to.
Actions
The gestures of surgery — retract, dissect, clip, divide — recognized as they are performed.
Phases
Where the procedure stands in its expected flow, and when it departs from that flow.
Spatial relationships
What is near what, what is between, what is behind — the geometry that makes a moment safe or critical.
Temporal relationships
What came before and what it implies: understanding accumulates across the whole procedure.
Context
The procedure being performed, the phase it is in, and the evidence gathered so far — the frame that gives meaning.
Evidence
Every interpretation is grounded in observations that can be pointed to, reviewed, and challenged.
The same moment.
A different meaning.
Below, the observation never changes: an instrument approaches a structure. Change the context around it, and watch what the observation means change with it.
The observation · fixed
Instrument approaches a tubular structure.
The context · yours to change
Routine approach
Early in exposure, with a confirmed structure and no complicating history, this motion is exactly what the procedure's flow predicts.
Illustrative interaction, simplified for clarity — not output from the live system, and not clinical guidance.
This is why surgical intelligence cannot be a single model: meaning lives in procedure, phase, history, and evidence — so the architecture has to carry them.
06 · From perception to decision
One continuous loop, running across the procedure — every step inspectable, every conclusion traceable back to what was observed.
Always advisory · Never autonomous
Intelligence you can take apart.
Most AI systems ask to be believed. Trigone is architected to be checked. Underneath the intelligence sits a knowledge architecture with rules about what counts as evidence, where knowledge comes from, and who has the authority to admit it.
Anatomy of a claim
Evidence, not vibes
Every claim the system makes is grounded in observations it can point to. If the evidence is thin, the claim says so.
Provenance, end to end
Knowledge carries its origin with it — where it came from, how it was derived, and under what license and authority it entered the system.
Certainty, made explicit
Confidence is a first-class quantity: measured, calibrated, and reported — never implied by a confident tone.
Governance, by design
What the system is allowed to learn, claim, and surface is governed by explicit rules with human authority at the top — not by whatever the model absorbed.
AI proposes.
Humans decide what is true.
In Trigone, no model output becomes trusted knowledge on its own. The boundary between automation and human authority is built into the architecture — watch it enforce itself.
Proposals
proposed claim
Phase transition recognized
confidence 0.84 · evidence attached
Trusted knowledge
AI proposes. The system generates a claim with its evidence and calibrated confidence attached.
This is the line that makes everything else trustworthy: automation accelerates the work, and people decide what becomes knowledge.
Validated before claimed.
Trust in surgical AI has to be earned in a specific order: engineering rigor first, structured human validation next, clinical evidence last. We hold ourselves to that order — and say plainly where we are on it.
Where we stand
Trigone is an investigational platform. Its engineering foundations are built and systematically verified; expert human validation is being prepared; it is not a cleared medical device and not clinically validated. We are building toward validated surgical intelligence — and we will not describe it as more than it is.
Deterministic by design
The same case, replayed, produces the same result — a precondition for debugging, audit, and regulatory scrutiny.
Benchmarked, not demoed
Capabilities are measured against defined benchmarks with frozen protocols, including the failure cases.
Calibrated confidence
A confidence score is only useful if it means what it says. Calibration is measured and corrected, not assumed.
Structured human validation
Expert review is designed like an experiment: blinded where it matters, contamination-controlled, with authority to say no.
Built in horizons.
Claimed one at a time.
Each horizon stands on the one before it, and each is labeled for what it is: what we are building, what is in development, and what remains vision.
Surgical workflow understanding
Perceiving instruments, anatomy, and actions, and recognizing how a procedure unfolds — the measurable foundation everything else stands on.
Context-aware surgical intelligence
Connecting perception to procedure, phase, history, and evidence — so the system understands what a moment means, not just what it contains.
Decision support
Advisory intelligence at the point of care: the right context surfaced at the right moment, with explicit confidence — always under the surgeon's control.
Surgical intelligence infrastructure
A governed platform for surgical knowledge — supporting training, quality, and research across institutions, with provenance and human authority intact.
Trigone
Surgery is one of the most complex environments in medicine.
We are building the intelligence to understand it.
For surgeons, hospitals, researchers, and partners who believe the operating room deserves better intelligence.