One person, many agents, one memory

Someone goes on vacation, an agent gets swapped, a better model shows up next month. In every one of those cases the work has to continue, and it continues the way it always has: someone re-explains the context, the new agent starts from zero, the old decision gets re-litigated as if it had never been made.

The ValorBrain bet is easy to state: a company memory is either shared or useless. It lives in one place, with version, source and permission, and it gets delivered to whoever asks, person or agent. When something changes, the previous version is superseded, not deleted.

The product is built inside that thesis: ValorBrain's own development runs on ValorBrain. The decisions, the handoffs between agents and the bug lessons the team records in the engine are the same structures customers use. If the memory couldn't hold our work, it wouldn't survive yours.

Valor Digital has run B2B products in Itapema, Santa Catarina, Brazil, since 2024, with AI agents in the daily flow of code, operations, marketing and legal. The problem that kept repeating: every session started from zero. A decision made in Claude never reached Cursor. Handoffs were copy-paste. Context died when the chat closed.

The open-source space has plenty of good work: QMD, A-MEM, MAGMA, Honcho, MemPalace, Hindsight, supermemory, mem0 and others each solve part of the problem, some very well. None of them closed the whole product: real multi-tenant isolation, hosted MCP, governance with audit, OAuth connectors, and an account a company can actually buy.

ValorBrain is the tool our own work was missing, opened to other teams. It integrates patterns published by those projects, with an honest feature-by-feature comparison in /comparison.

ValorBrain is a Valor Digital product. Gus is the founder: he designs, writes and maintains the product with a fleet of AI agents operating on the same memory the product sells. A small team is a fact, not a phase; the cost shows up in the roadmap and the payoff shows up in coherence.

The code is closed, for security. The roadmap is public, version by version, at /changelog. Benchmarks, method and caveats live in /benchmarks, and the state of security controls, including what has not yet been verified, in /security.