Product
Learns · Knowledge Compiler Briefs · Context Engine Yours · Trust LayerSurfaces
Enterprise · Soma People · Self Family / team · Self+ M2M · Axis Protocol · NexusIP
Patent · P202631047 MVE · Minimum Viable Exposure .aix format ↗ Research Contact hello@thegeometrical.com02 · Context Engine
Every AI gets exactly the context it needs. Nothing more.
Hybrid retrieval across three kinds of memory, the reasons behind each decision, the procedure your team follows — trimmed to a budget and sent at Minimum Viable Exposure to any model or agent.
Q: “Where did the last conversation leave off — and what’s still pending?”
Nearly 2× the right answers.
On 21× less context.
Real benchmark, not a mock-up · 72% vs 39% correct · 1,684 vs 35,282 tokens per question · vs. the best raw retrieval (top-128) · same model, same questions · single-user graph running since July 2026 · 12 questions × 3 runs · Claude Sonnet 5 answering only from the given context · blind LLM judge · official API token counts · Oct 2026. Graph and model windows are illustrative.
How it works
From the moment an agent asks to the moment it gets an answer it can trust.
“What should I know before the Banca Ribera call?”
trace.log
context package — tok
Illustrative trace with fictional data. Token figures from our October 2026 benchmark (35,282 raw vs 1,684 delivered).
What’s inside
Briefs is what any agent sees of your graph: the least it needs, structured, under your rules.
05 · Procedures
Learned know-how
When your team repeats the same steps, Geometrical turns them into a versioned procedure. Any agent — and every new hire — follows the latest version.
Live
06 · Three memories
Hybrid retrieval
Episodic for what happened, semantic for what’s true, procedural for how it’s done — searched by keywords, meaning and graph importance together.
Live
07 · Exposure
The least data a model needs
Each call gets only what the task needs, inside a token budget — and its Exposure Ratio is recorded, never the sensitive values.
Proposed open standard · CC BY 4.0
08 · Any model
Neutral by design
Claude today, another model tomorrow. The graph is served over MCP or API and exported in .aix, an open format — switching models never resets what your AI knows.
Live
Why own a graph
Without a graph, every AI session re-sends what your team already explained — and still gets it wrong more often. Put in your numbers.
21 working days / month. Change the price to your provider’s.
Per question: 35,282 vs 1,684 tokens · 39% vs 72% correct — measured on a single-user graph (12 questions × 3 runs), same model, Oct 2026. Raw = best raw retrieval (top-128). Input-context spend only.
An imperfect graph is worth more than none — and it gets stronger with time. Illustrative curves; the 21× ratio is measured.
| Apple | Microsoft | Your graph | ||
|---|---|---|---|---|
| Lives in | Apple devices & iCloud | your Microsoft 365 tenant | your Google account | your perimeter |
| Remembers | what happens in Apple’s apps | what happens in Microsoft’s apps | what happens in Google’s apps | all of it, across tools |
| Answers with | Apple Intelligence & partners | Copilot | Gemini | any model, any agent |
| Built for | one person | one tenant | one account | people, teams, organizations |
| Leaves with you | within Apple’s ecosystem | within your tenant | within your Google account | .aix · open format |
| Who decides who reads | Apple | Microsoft & your admin | you, per agent (Nexus) |
They’ll each remember a slice of you. Only you can own the whole graph.
Based on public materials, Oct 2026.
Results
72% vs 39%
Same model, same questions. Geometrical’s compiled context against the best raw retrieval.
Internal benchmark · single-user graph · Oct 2026
Learn more →21×
1,684 tokens instead of 35,282 — and more answers right, not fewer.
Internal benchmark · single-user graph · Oct 2026
Learn more →39×
2,332 instead of 90,700. The saving shows up on every model bill.
Internal benchmark · single-user graph · Oct 2026
Learn more →1.9 s
Median, with no generative model in the loop.
Measured · Oct 2026
Learn more →Next
Yours. →It lives inside your perimeter. Every agent asks permission. We never see what’s in it.