Privacy policy
How Pcnaid Inc. handles information on this website and in the hosted Pcnaid Memory OS service — written against what the software actually does, not against a template.
- Effective 16 August 2026
- Pcnaid Inc.
- No trackers on this site
The short version
This website sets no cookies, runs no analytics and loads nothing from a third party. The Memory OS service stores what you and your agents write to it, isolated per tenant by database row-level security and recorded in a tamper-evident audit log. We do not sell or share personal information, we run no advertising, and we never train a model on your tenant content.
This is a plain-English summary for orientation. It does not replace the policy — the numbered sections below are the terms that apply.
What this policy covers
Pcnaid Inc. (“Pcnaid”, “we”, “us”) publishes Pcnaid Memory OS, durable memory infrastructure for AI agents. This policy is effective from 16 August 2026 and covers two things: this marketing website, and the hosted Memory OS service at memoryos.pndsvcs.com.
It does not cover an installation you run yourself. The software is MIT licensed and self-hosting is unrestricted; when you run your own instance, your Postgres database holds your data, we are not in the path, and we hold nothing. In that case you are the operator and this policy describes none of it.
It also does not cover the products you build on top of Memory OS. If you reached this page because an application you use stores its memory here, the operator of that application is the organisation you need to contact — see your rights below.
Two different things: the website and the service
They behave very differently and it is worth separating them before anything else.
This website is a static site. Every page is
pre-rendered HTML served from a content delivery network. It sets no
cookies, runs no analytics, embeds no third-party media, loads no
fonts from a CDN — the typeface is served from our own origin — and
runs no tag manager, pixel or session recorder. The only thing it
writes to your browser is a single
localStorage key, memoryos-theme, holding
the word light or dark. It is never sent to
a server. The
cookie and local-storage policy lists
every item in full.
The service is an authenticated application with a database behind it. It holds account identity, whatever your agents store, billing state and an audit trail. Everything from the next section onwards is mainly about the service.
Information we collect
From this website
Nothing beyond the standard server logs that any web host keeps in order to serve a file: the requesting IP address, a timestamp, the path requested, the HTTP status and the user-agent string. These are generated at the CDN edge that serves the site, are used for delivery and abuse prevention, and are not assembled into a profile, not linked to an account and not sold. We do not operate an analytics product on this site, so there is no visitor record for us to look up.
From the Memory OS service
- Account identity, via Auth0. The subject identifier, email address, the display name your identity provider supplies, and the tenant and role claims that authorise the session. Memory OS validates the token signature against Auth0’s JWKS and refuses any request whose token carries no tenant claim.
- Tenant content — whatever you choose to store. Memories, bitemporal facts, Fabric entries, imported sessions and transcripts, knowledge-base sources and the documents fetched from them, and text derived from media such as transcripts, OCR output and captions. We do not select this content; your agents and your people write it.
- Billing identifiers, via Stripe. Your Stripe customer and subscription identifiers, plan, quota counters and invoice state. Pcnaid does not store card numbers — card details are entered on Stripe’s own hosted checkout and held by Stripe. We never see them.
- Support tickets. The subject, message body and attachments you send through the in-product support desk or by email, plus the tenant they came from.
- Audit metadata. Every memory operation appends an audit row recording who did what, when, against which record, with the request ID that ties it to the API call. Each row carries the hash of the row before it, so tampering is detectable.
There is content we ask you not to put in: passwords, API
keys and private keys, full payment card or bank credentials,
government identifiers such as national insurance or social security
numbers, and medical records or other regulated sensitive categories.
Our security notes say this plainly, and the product gives you
DO_NOT_STORE: for the one-off case where sensitive text
passes through a conversation and must never be written.
How we use it
- To operate the service — storing, indexing, retrieving and returning your own data to you.
- To authenticate you and to enforce the tenant boundary on every query.
- To meter usage against your plan, take payment and issue invoices.
- To answer your support requests and to notify you about incidents, security issues and material changes.
- To investigate abuse, fraud and threats to the integrity of the service.
- To keep the audit chain, which exists so that you can prove what happened.
- To improve the product using aggregate operational measurements — latency, error rates, queue depth, quota pressure. Not the content of your memories.
Legal bases
If you are in the United Kingdom or the European Economic Area, we rely on the following bases where we act as a controller.
- Performance of a contract — providing the service you have subscribed to, and supporting it.
- Legitimate interests — securing the service, preventing abuse, keeping an audit trail, and understanding aggregate operational load. We have balanced these against your interests and consider them proportionate.
- Legal obligation — retaining billing and tax records for the period the law requires.
- Consent — where we ask for it, such as when an operator chooses to configure a third-party AI provider that will receive content. Consent can be withdrawn by removing the configuration.
Where we act as a processor on your instructions, you determine the basis for processing the content you store, and you are responsible for having one.
Controller and processor
The distinction matters, because the two halves of the service are governed differently.
For tenant content, Pcnaid is a processor. You decide what your agents write, why, and for how long. We process it on your documented instructions — which in practice means the API calls your systems make — and for no independent purpose of our own. We do not mine it, sell it, or use it to train any model. We will assist you in responding to a data-subject request that concerns content in your tenant, and on termination we will return it as an export or delete it as you direct.
For account identity, billing, support correspondence and our own audit of the service, Pcnaid is a controller. We decide that we need an email address to authenticate you, a Stripe customer record to bill you and an audit trail to run the service safely.
We will enter into a written data processing agreement, including standard contractual clauses where they are required, on request to support@pcnaid.com.
Sub-processors
These are the categories of third party that may process data on our behalf, and the specific providers currently used in each.
| Category | Provider | What it processes |
|---|---|---|
| Authentication | Auth0 | Login, token issuance, the identity claims on the session. Auth0 sets its own cookies at its own domain during login. |
| Payments | Stripe | Checkout, card data, subscriptions, invoices. Card numbers reach Stripe and never reach us. |
| Website hosting | Cloudflare Pages | Serves this static marketing site and generates the edge request logs described above. |
| Application and database hosting | Our cloud infrastructure provider | Runs the API and the Postgres database that holds tenant data. Named on request. |
| Error reporting | Sentry — optional | Stack traces and request context, only when a DSN is configured. A scrubber redacts cookies, authorisation headers, tokens, email addresses and phone numbers before an event leaves the process. |
| Tracing | OpenTelemetry — optional, off by default | Span timings to whatever collector the operator points it at. Disabled unless explicitly enabled. |
| AI inference | OpenAI, Anthropic, Google Gemini or any OpenAI-compatible endpoint — optional | Only the content sent for embedding, summarisation or chat, and only when the operator configures a provider. |
| None | The application sends its own mail over SMTP. There is no hosted mail vendor in the path. | |
| Support desk and status page | None | Both are in-repository features of the product itself. No customer payload is sent to a support or status vendor, and a continuous-integration check enforces it. |
Two points about the AI providers, because they are the ones that see content. First, they are engaged by configuration: no provider receives anything until an operator sets a key. Second, they are avoidable entirely. Memory OS ships deterministic fallbacks — extractive summarisation and a local embedding model — so a tenant can run with no external AI provider at all and nothing silently stops working. Where you bring your own key, the provider processes that content under your agreement with them.
International transfers
Pcnaid Inc. is a United States company and the service is operated from the United States. If you are in the United Kingdom or the European Economic Area, your data will be transferred to and processed in the United States.
Where such a transfer requires a safeguard, we rely on the standard contractual clauses, which we will execute as part of the data processing agreement described above. We do not claim certification under any transfer framework. If you configure a third-party AI provider, content sent to it may be processed in whatever regions that provider operates, under its terms.
How long we keep it
- Tenant content is kept until you delete it. You can remove individual records with the preview-then-confirm forget flow, remove content by API, or delete the whole tenant.
- Content marked
DO_NOT_STORE:is never written in the first place, and no pseudonymisation mapping is created for it either. - Tenant delete is a soft delete followed by a purge. Deleting a tenant marks it deleted and sets a purge date a configurable number of days later — thirty by default. A scheduler runs every six hours, finds tenants whose purge date has passed and hard-deletes their rows across every tenant table: memories, facts, knowledge-base sources and documents, the PII vault, the outbox, audit events, API keys, usage counters, settings, support tickets, idempotency keys and onboarding state. The grace period exists so an accidental deletion can be reversed; once the purge runs, the rows are gone.
- Audit events live in the tenant tables and are purged with the tenant. Until then they are retained deliberately — an audit log you can edit is not an audit log.
- Billing records are retained by Stripe and by us for as long as tax and accounting law requires, typically seven years, independently of tenant deletion.
- Support correspondence is stored with the tenant and purged with it.
- Edge request logs for this website are retained on the host’s standard short cycle and are not archived by us.
How we protect it
These are controls that exist in the code, not aspirations.
- Row-level security, enabled and forced on every
tenant table. The tenant is set per transaction as a session
variable and the policy is evaluated by Postgres, so a query that
forgets to filter still cannot cross the boundary. The production
runtime role is created
NOSUPERUSER NOBYPASSRLS, and continuous integration asserts that the role in use is not an owner, a superuser or aBYPASSRLSrole — because those bypass policies even when a table forces them. - A hash-chained audit log. Each audit row stores the hash of the previous row and a hash of itself, so removing or editing an entry breaks the chain. A verifier walks the chain and reports the first break.
- A PII pseudonymisation vault. Email addresses and
US-format phone numbers are replaced with placeholders
(
EMAIL_1,PHONE_1) before the text is embedded, stored or injected into a model’s context. The real values live in a tenant-scoped vault protected by its own policy, substituted back only at display, and revealing a mapping requires owner permission. Detection covers email addresses and US phone formats only — it is not a general-purpose PII classifier, and we say so rather than implying more. - Encryption of stored credentials. Provider keys you bring are encrypted at rest with Fernet. API keys are stored as hashes, shown once at creation and never retrievable afterwards.
- Constant-time token comparison for admin and alert tokens, so a comparison cannot be timed to recover a secret.
- SSRF defences on knowledge-base fetching: scheme and domain allowlisting, DNS resolution checked before the request, private, loopback, link-local, multicast and reserved ranges blocked, a response size ceiling, and a resolve-once pinned-IP fetch that mitigates DNS rebinding.
- Request-size ceilings, allowlisted CORS and rate limiting. The rate limiter is in-process and does not coordinate across replicas; it is a guard rail, not a distributed quota.
No set of controls is absolute. If you believe you have found a weakness, please report it privately to security@pcnaid.com. We aim to acknowledge within seventy-two hours.
What we are not certified for
Stated plainly, because the alternative is letting a security section imply something it has not earned:
- Pcnaid holds no SOC 2, ISO 27001, HIPAA or PCI certification for Memory OS, and has published no third-party penetration test.
- Memory OS is not a system of record for regulated medical data. Do not store protected health information in it.
- Memory OS is not in scope for payment card data. Do not store card numbers in it. Card handling is Stripe’s, on Stripe’s infrastructure.
If your use case requires a certified environment, self-hosting inside one is the honest answer, and the MIT licence makes it free to do.
What we do not do
- No advertising, anywhere, of any kind.
- No selling of personal information, and no sharing of it for money or other valuable consideration.
- No cross-context behavioural advertising, and no sharing for that purpose.
- No training of any model — ours or anyone else’s — on your tenant content.
- No analytics, session recording or fingerprinting on this website.
- No reading of your memories by our staff except where you ask us to in support, or where we must to investigate a specific security incident. Such access is written to the audit chain like everything else.
Your rights, and how to exercise them
Depending on where you live, you have some or all of the following rights: access to the personal data we hold about you, rectification of it if it is wrong, erasure, portability in a machine-readable form, restriction of processing, objection to processing carried out on the basis of legitimate interests, and withdrawal of consent where processing rests on it. You also have the right to complain to your supervisory authority.
Write to support@pcnaid.com. It is the correct address for every privacy and data-subject request; there is no separate privacy mailbox to find. We will acknowledge promptly and respond within thirty days, and we may need to verify that you control the account before acting.
If you are an end user of a product that someone else built on Memory OS, that organisation is the controller of your data. Contact them first. Where they ask us to act, we will assist them.
Two rights have concrete mechanisms in the product:
- Portability — the tenant export produces a ZIP archive of the tenant’s rows across every tenant table as CSV and JSON. It can be written to local disk or to object storage with server-side AES-256 encryption. Ask us and we will run it for your tenant; if you self-host, it is a script you run yourself.
- Erasure — the tenant delete endpoint marks the tenant deleted, sets the purge date, and the six-hourly scheduler hard-deletes it when the grace period expires. For individual records, the forget flow previews exactly what will be removed and requires you to confirm by identifier; a bare instruction to forget something is refused by default, precisely so that a stray sentence in a transcript cannot delete your memory.
Children’s data
Memory OS is a developer product sold to organisations. It is not directed at children, and we do not knowingly collect personal data from anyone under sixteen. If you believe a child has provided us with personal data, write to support@pcnaid.com and we will delete it.
Automated decision-making
We carry out no automated decision-making that produces legal effects or similarly significant effects on a person, and no profiling for marketing or advertising.
The service does rank and filter records during retrieval — fusing search results, scoring trust, suppressing facts that a later fact has superseded. That is ranking within your own data to decide what a model sees, not a decision about a person, and the source manifest returned with every context pack tells you exactly which records were selected.
Changes to this policy
When this policy changes we will update the effective date at the top. For a material change — a new category of data, a new sub-processor that receives content, a change in the basis on which we process — we will notify tenant owners by email before it takes effect. Previous versions are available on request.
How to contact us
Pcnaid Inc. publishes Pcnaid Memory OS. These are the only three addresses we operate, and all three are read by people.
Related: terms of service · cookies and local storage · accessibility.