About
Persistent memory for the AI agents your enterprise actually runs.
This page is the long-form version. The home page covers what Mycelium does and how it ships. This page covers why the name exists, what we believe, and who is building it.
The name
A mycelium is the underground network of fungal threads that connects every tree and plant in a forest. Nutrients move along it. Signals move along it. Memory of what the ecosystem has experienced moves along it. The forest above ground is the visible outcome; the network below is the load-bearing infrastructure.
We picked the name because the metaphor is the architecture. Persistent shared memory connects every agent, every system, every operator across an enterprise. The agents and dashboards your business sees are the visible outcome. The memory layer underneath is the load-bearing infrastructure that makes the rest hold together.
What we believe
Most AI agents fail not because the models are weak, but because they have no memory of what your business has decided, who is involved, why a policy exists, or what was tried last quarter. Without persistent shared memory, every agent is a stranger to your company every time it wakes up.
The fix is not bigger context windows. The fix is a memory layer that sits between the agent and the enterprise: typed, queryable, version-controlled, portable across model vendors, reliable enough to plug into procurement.
We build for the company that has already adopted AI agents internally and now needs them to actually act inside the business. The work is not about model selection. The work is about the layer that makes the model usable on Tuesday.
Open core, on purpose
The memory layer ships open source under MIT. The pattern, the schema, the typed graph, the conventions for write-back and resolution: anyone can clone the repository, install the layer on a single machine in two hours, and run the same primitives locally that we run for enterprise.
The runtime that operates the memory layer across an enterprise (multi-tenant authentication, audit, real-time event streams, webhook reliability, single-tenant or on-premise deployment) ships under commercial license. Same shape as Vercel and Supabase: open core that teaches the pattern, paid runtime that carries it across teams.
We picked this shape because it forces honesty. If the open-source layer does not actually work, no one runs it. If it does, the path from a single operator to an enterprise install is continuous, not a rebuild.
Reliability is a discipline, not a claim
Every release passes 59 automated checks before it ships. End-to-end integration tests run on every code change. The reliability manifesto and recovery runbook are public in the GitHub repository. The release does not go out if any check fails.
We chose to publish the reliability artifacts because procurement teams ask the same questions every time: how do you test, what happens when something breaks, what does the audit trail look like. The answer is in the repository, and the answer is the same whether you are a single operator or a Fortune 100 vendor risk team.
The operators behind it
Two founders. Built by people who have run the kind of business that buys software like this, not by researchers describing what should be possible.
Adelaida Diaz-Roa
Founder
Seven companies, two exits. Co-founded Villy Customs at 18; the company closed an investment from Mark Cuban and Barbara Corcoran on Shark Tank. Pawliday Inn was acquired by a national dog hotel brand entering the Dallas market. Success Magazine 30 Under 30.
Leads product, architecture, and install delivery.
Sergio Pérez Barón
Co-Founder
Kellogg MBA. Ten-plus years at the World Bank in operator-side roles spanning emerging-markets program management, finance, and procurement strategy.
Leads revenue, partnerships, and enterprise sales.