Context Intelligence
Give every AI agent the context it needs for the task, including relevant history, relationships, boundaries, changes, and human decisions, before it acts.
The product
What GammaLex read, what it picked, what it found wrong, and why it stopped the agent.
What happened
Task
Prepare for today’s visit.
GammaLex read 327 permitted records, chose the 14 this task depends on, and stopped the agent because one of them is not there.
What it found
BlockedWhy it stopped
MISSING
It has come up at two visits and was never written down. The provider can still ask — the patient is in the room.
What it read
14
records, out of the 327 it was allowed to open
Selected from this patient’s chart, the current episode, active medications, and the decisions this provider already approved.
How GammaLex works
GammaLex connects to the information an agent is permitted to use, assembles what matters for the task, and checks whether the context is ready before the agent proceeds.
01
What the agent is about to do, who is acting, and for whom.
What the agent asks for
get_context(
task: "Prepare for today's visit",
actor: treating provider,
scope: this patient,
)02
Rank the permitted records against the task — then follow the relationships between them, so a record reached only through another one still arrives.
What it picked — 14 records of 327
+ 11 more · 313 held at the boundary
03
Report what is missing, conflicting, changed, uncertain, or out of bounds — and say nothing rather than guess.
What it found — and stopped
04
Hand back the sources, the reason each one was chosen, and the verdict.
What the agent gets back
Core capabilities
Developed and tested against real work — patent intelligence alongside a practicing attorney at KPPB Law, and clinical intelligence through GammaLexCare.
For one job, gather the records it depends on, how they connect, what came before, and the reason each was picked.
Applications
GammaLexLegal and GammaLexCare are applications of GammaLex Context Intelligence, and the proving grounds where it is developed and tested against real work.
Independent and concierge practices
Show up to every visit already knowing.
GammaLexCare carries prior visits, care plans, patient-reported changes, and provider-approved decisions into the conversation happening now — and says what is still missing while there is time to ask.
What it connects
Patent work
Context Intelligence for patent work.
GammaLexLegal connects the pending claim set, cited prior art, prosecution history, and amendments around the patent task being performed.
What it connects
OpenContextually
In developmentGammaLex's open-source context layer answers one question for a coding agent: which files should it read before it starts working on this task? It never edits anything. It hands the agent a reading list.
It is the same idea GammaLexLegal and GammaLexCare apply to a patent file and a patient chart — proven first on code, where the connections between things are already written down.
Every file it is allowed to open is scored against the job — the name first, then what is defined inside it, then the text. Saying a word two hundred times does not win.
Having found one file, it follows what that file depends on — and pulls those in too, even when they never mention what you searched for.
Not a list of filenames, which makes the agent go read everything. Not whole files, which fill up its memory. Just the lines that earned the file its place — with anything secret stripped out.
Its checks say nothing rather than guess. On four real codebases, neither one spoke up — which is right, not a gap.
GammaLex did not begin with a product category.
It began with a standard I was no longer willing to compromise.
If AI is going to influence decisions that matter, it should not act on whatever information happens to be available. It should understand what came before, what changed, what remains uncertain, and where human judgment belongs.
That is why we are building Context Intelligence at GammaLex.
Build with GammaLex or help us test Context Intelligence in real work.