Running in production
The whole-company layer, running at n=1.
Trovano is one governed brain with domain-specialized agents on top of it. This page is not a mockup. Every number below is queried live from the system as this page renders, including a real construction company running as its own partition.
84,663
entities in the brain
The fleet
Each Core is a specialized agent with its own memory, its own governance hooks, and its own partition of the shared brain. They read across each other on request; they never write into each other.
Lifeorg_id 1
Personal operations, health, relationships, projects
39,681 entities
88,833 edges
3,392 evidence
69 sessions
34 registered governance hooks · 3 review agents
Businessorg_id 2
This company: customers, pipeline, strategy, hiring
12,381 entities
16,160 edges
170 evidence
22 sessions
28 registered governance hooks · 3 review agents
Schoolorg_id 3
OSU coursework, deadlines, assignments
15,253 entities
24,405 edges
163 evidence
32 sessions
28 registered governance hooks · 3 review agents
Financeorg_id 4
Cash, runway, positions, financial decisions
4,518 entities
11,108 edges
43 evidence
7 sessions
28 registered governance hooks · 3 review agents
Constructionorg_id 5
A real construction company's books and operations
12,830 entities
11,873 edges
1 evidence
9 sessions
28 registered governance hooks · 3 review agents
What the brain has actually learned
The brain does not just store documents. It compiles what happened into typed knowledge, which is what lets it answer with what is currently true instead of what used to be.
Source
27,040
files and sessions it learned from
Topic
18,512
subject clusters
Entity
12,473
people, orgs, systems
Decision
9,988
choices made, with the why
Incident
7,047
things that went wrong
Rule
5,266
constraints it now enforces on itself
Lesson
2,709
corrections turned into behavior
Tool
1,285
tools and frameworks used
How it stays honest
Governance runs as code, not discipline. Each Core registers 28 or more governance hooks that gate the agent's own behavior: they block claims about system state that were not verified by a read this turn, block "I did X" when no write happened, block time claims with no clock call, and force a memory query before assertions about past decisions.
Provenance is bi-temporal. Every entity carries valid_from / valid_until and a superseded_by pointer, so the system can answer what was true at a past date, not only what is true now.
Retrieval is hybrid. Vector similarity, full-text search, and graph traversal are fused, so recall does not depend on guessing the right keyword.
What this proves, and what it does not. It proves the architecture works and that one person can build and operate it across five domains, including a real company. It does not prove multi-tenant operation, extraction on other companies' messy data, or anything about scale beyond a single operator. One user. No paying customers. Evidence spans 2026-05-17 to 2026-07-25. Counts and structure are shown here; no memory contents, names, or financial figures are exposed.