Pcnaid Memory OS
Get started

Infrastructure we run ourselves

Pcnaid Inc. builds operating systems for businesses that outgrew spreadsheets and refuse enterprise pricing. Memory OS is the memory layer underneath that line — built because our own agents kept forgetting things.

Why this exists

We build and operate a number of products, and increasingly the people working on them are agents. Agents that read a codebase, make a decision, do the work, and then — three days later — make the opposite decision, because nothing remembered the first one.

The available answer was a vector store. It was not enough. A vector store can tell you a sentence looks like your question. It cannot tell you whether the sentence is still true, who decided it, what it replaced, or whether the document it came from should be trusted at all. Those turned out to be the questions that mattered.

So Memory OS is a system of record rather than an index: separate layers for separate kinds of knowing, time on every fact, evidence on every decision, and a hard structural rule that retrieved records are data and never instructions. We run it in production against our own work, which is the main reason the unglamorous parts — the outbox, the audit chain, the dead-letter console — exist at all.

The state of it, plainly

Version
0.3.6
Licence
MIT, full permissive grant
API surface
141 endpoints, 32 route families
Agent tools
18 over stdio MCP
Backend coverage
87.07% branch, against an 85% gate
Frontend tests
77 passing across 27 files
Certifications
None. See below.

How we build

Five rules we actually enforce, including the one that made this site harder to write.

Ship the whole vertical or do not ship it
A feature is finished when the interface, the state, the API, the service, the data, the schema, the seed and the tests all exist and have been run. A stub that returns the right shape is not a feature, and we do not count it as one.
Fix the cause, not the symptom
When something breaks, the question is which invariant was missing, not which line to patch. A fix that leaves the same class of bug possible somewhere else has not finished.
Evidence, never "should work"
Every claim on this site was checked against the source or against the running service. Where we could not check something, we say so on the page rather than rounding it up.
Say the unflattering part out loud
This site tells you that self-serve signup does not exist, that our benchmark has two cases, that the cognitive tier writes to a file rather than Postgres, and that several features are heuristic rather than model-driven. Anyone evaluating infrastructure finds these things eventually. Better here than in production.
No borrowed credibility
No customer logos we have not earned, no certification badges we do not hold, no partner marks we lack written clearance for, and no benchmark numbers dressed up as something they are not.

Security, and what we do not claim

Memory OS enforces tenant isolation in the database rather than in application code, keeps a tamper-evident hash-chained audit log, pseudonymises personal identifiers before they reach an embedding, and runs no third-party support, status or email vendor — a rule a CI check enforces on every commit.

What it does not have is any external attestation. There is no SOC 2 report, no ISO 27001 certificate, no HIPAA or PCI position and no third-party penetration test. Our own security notes explicitly say regulated medical data should not be stored in the service. If your procurement process requires one of those, we are not there yet, and we would rather you knew now.

Found something? Write to security@pcnaid.com. We aim to acknowledge within 72 hours and to fix critical issues within 14 days when the fix is under our control.

How isolation is actually enforced

Where to reach us

Keep reading

Give your agents a memory you can audit

Run the whole system on your own hardware under the MIT licence, or ask us about hosted access. Both start from the same place.

Get started Read the API docs