Back to the exchange

Privacy

NomosX Privacy Policy

Effective November 10, 2026.

Nomos Arc, Inc. ("NomosX," "we," "us"), a Delaware corporation, operates nomosx.io, the NomosX plugin, and the NomosX earnings program (the "Service"). This policy explains what we collect, why, and your rights. Nomos Arc, Inc. is the data controller for the processing described here.

Contact: privacy@nomosx.io

1. The short version

Start here, because this is the part you most need to know. Three things derived from your sessions are sent to NomosX. Nothing else derived from your sessions is.

  1. 01
    Interest signals. The matching categories your work falls into, with weights; a short slug and category naming the thing you are currently evaluating; and the slugs of vendors you have rejected. These are category labels and short names taken from the notes on your machine. They are never text from your conversation, and they are what the sponsored result on your status line is matched against.
  2. 02
    A short task summary, for each match made inside a session. When the Plugin's commerce tool runs in a conversation, your assistant writes a line or two describing the task, for example "user is selecting an error-tracking tool for a Node.js service", and assigns the task a category, which is sent with the summary. The summary is capped in length and filtered when we receive it. We call these summaries, each with its category, derived artifacts.
  3. 03
    Your profile wiki, while the program is on. The Plugin keeps a set of short notes on your machine about your interests, the tools you use, and what you are currently evaluating: one page per interest, per buying intent, and per vendor you have used or rejected, up to 500 pages of at most 64KB each. While the program is on, that set replicates to our servers: the page text as your machine wrote it, filtered on your machine first, together with the graph of which vendors you are weighing, chose, or rejected. This is the richest thing we hold about you, and it is not the thing that selects the sponsored results you see. Section 4a states the four purposes it actually serves today, and Section 3 says what is filtered out of it first.

On the Claude and ChatGPT connectors, which run no NomosX software on your machine, only the second of these is sent: the task summary, with the category your assistant assigns to it.

Your raw conversations are never transmitted to NomosX. What we never receive is the conversation itself: your prompts as you typed them, the assistant's replies, your files, and your code. That is a real distinction and not a play on words, and here is what it rests on. Your profile wiki is not a copy of your sessions. It is written on your machine, from the prompts you typed, under a schema that tells the writer to record durable facts about your interests and choices and instructs it never to copy session text, people's names, file paths, or secrets. A deterministic filter then runs over each page on your machine before it is sent. So when this policy says we hold your page text verbatim, it means we store the page exactly as your machine produced it, byte for byte, not that the page is a transcript. The schema's instruction is an instruction to a language model rather than a guarantee, so if you find that a page recorded something it should not have, Section 8 is how to have it deleted.

The analysis that writes those notes runs on your machine, using your own AI provider account (for example your claude CLI), under your existing relationship with that provider. We supply the software and we determine what it looks for, so we are responsible for that processing, but the session text goes to your provider and never to a NomosX server.

Ads, earning, and personalization are one program: they all start when you install the Plugin and create an account, and there is no version of the product with ads and earning but without personalization. We are not going to pretend otherwise. Your attention and these signals are what you are paid for, and you receive 60% of the placement value we attribute to each billable impression, per impression, never contingent on clicking or buying anything.

How you stop it, and what each route actually reaches. There is one switch for the whole program. These routes are not equivalent, so they are set out separately:

  • The opt-out toggle in your account. It records the opt-out on your account and our servers enforce it for every machine you sign in from. A signed-in machine picks up the change within 24 hours, sooner when it next contacts NomosX on your account. This stops sponsored results, earning, and every kind of personalization together.
  • The Plugin's off command in Claude Code, Codex or Antigravity. It turns NomosX off on that machine only and changes nothing on your account: from then on that machine contacts NomosX only when you run a NomosX command there. Running the Plugin's install command in Claude Code, or its on command in Codex or Antigravity, turns that machine back on, unless your account is opted out.
  • Uninstalling the Plugin, or removing the NomosX connector. Uninstalling stops the Plugin on that machine only. The Claude and ChatGPT connectors have no off command; removing the NomosX connector in the app's settings removes it everywhere that connector is used: from every Claude app on that account, or from the ChatGPT desktop app and Codex on that computer. Neither changes your account settings, stops the program anywhere else, or deletes a replica we already hold.
  • Deleting a replica we hold is a separate act from stopping the program: use the erase control in your account, or ask us. Once you have been opted out for 90 days, we anonymize your data (Section 4a). Section 7 gives the retention period for each kind of record.

None of these costs you earnings you have already verified.

And what we never do. We do not sell your personal information and we do not share it for cross-context behavioral advertising. Advertisers never receive your identity, your conversations, your derived artifacts, or your wiki; what they receive is described in Section 5. Nothing tracks you outside the Service. Sensitive categories such as health, religion, politics, sexual life, ethnicity, trade-union membership and immigration status are excluded from our matching taxonomy, which means an advertiser cannot ask to reach people on those grounds and no such category is ever stored on your profile.

Running the payments requires identity, tax, and payout data, which is handled by Stripe.

Installing the Plugin and creating an account is the act that starts this: you are choosing to run a product whose stated and only purpose is to show you paid sponsored results and pay you for them. From then on the program is active until you stop it in one of the ways above.

2. What we collect

What stays on your machine (never transmitted to us):

DataHandling
Your raw assistant conversations (prompts, responses, code)Never transmitted to NomosX. Processed locally by the Plugin; the local wiki builder sends excerpts to your own AI provider under your existing account with them (Section 1)
Files, repositories, and terminal contentsNever accessed except as needed locally to build the notes and signals described in Section 3; never transmitted
Local wiki pages (the Plugin's machine-generated notes about your working context)Built on your machine. While the program is on, the page text is filtered on your machine and replicated to our servers (Section 4a), where it is stored encrypted at rest. The local copy is never altered by the filter

What we receive on our servers:

CategoryExamplesPurpose
Account dataYour email address, display name and account settings, and what your sign-in method needs. With email and password: a hash of your password, never the password itself. With Google or GitHub: your account ID with that provider, the email address, name and profile picture link it shares with us, whether it has verified that address, and the tokens it issues for the sign-in (from Google, also an ID token), which we store with your account and use only to complete the sign-in. We ask Google only for your basic profile and email address, and GitHub only for read access to your profile and email addresses. With a passkey: its public key, its credential ID, a count of its uses, and details of the authenticator (its type and model identifier, whether it is backed up, and how it connects); the private key never leaves your device, and nothing biometric is sent to usAccount operation and sign-in; notices
Interest signalsPer-user matching-category counters with weights; a rejected-vendor set; the short slug and category of what you are currently evaluating, sent as the query on the status-line surface (each on a rolling 90-day expiry)Ranking candidate sponsored results; honoring your vendor rejections
Derived artifactsShort, machine-generated summaries of your current working context, each sent with the category your assistant assigns to it, on the in-conversation surface (Section 3)Contextual ad matching
Impression and delivery dataWhich sponsored result was served, when, on which surface; render confirmation, where the surface supports it; label displayBilling advertisers; calculating your earnings; audit
App and versionThe name of the app the Plugin is running in and the Plugin's version, sent with every request the Plugin makes and recorded with each request for a sponsored result; where the app says so, how it displays results (Section 3)Showing each sponsored result in a form the app can display; requiring an update when a version is too old to serve; recording which app and version each sponsored result was served to
Click data (if you click)The fact that a sponsored result was clicked; the click routes through our redirect with a click identifierAttribution of conversions to advertisers; fraud prevention. Clicking never affects your payout
Payout and tax dataName, tax forms (W-9/W-8), payout status. Bank and identity documents are collected and held by Stripe, not usPayouts; tax reporting; sanctions screening
Fraud and security signalsRate and deduplication counters on impressions; impression billability codes (the recorded reason an impression did or did not count); render-confirmation status; a per-install OAuth client credential identifying your Plugin install; and, for each signed-in session, the IP address and the browser or device description (user agent) it was started from, recorded by our identity serverDetecting invalid activity; protecting the exchange
Replicated wiki (while the program is on: Section 4a)Your wiki page text (encrypted at rest) and its intent-edge graph of vendors you are evaluating, chose, or rejectedGiving you access to and export of what we hold; audited operator review; aggregate reach counts; compiling a short structured summary of it (Sections 4a and 5)
Support and communicationsMessages you send usSupport
Access requests and the contact form on nomosx.ioThe email address and role you submit and, where you give them, the topic you choose and the message you writeAnswering you: we use the topic and the message only to reply to you. Resend, our email provider, delivers them to our team inbox as a processor; the acknowledgement it sends you never repeats your message
Website dataSee Section 9 (cookies and local storage)Site operation and analytics

We do not collect advertising identifiers or browsing history, we do not buy data about you from data brokers, and we receive nothing about your activity in apps or services the Plugin does not run in. Within the assistants it does run in, note that your local wiki is built by digesting your sessions in Claude Code, in Codex (its CLI, IDE extension and desktop app, and the ChatGPT desktop app, which keeps its sessions in the same place as Codex), and in Antigravity, in each only where the Plugin is installed and enabled, so while the program is on the replica can reflect activity across those assistants.

3. What leaves your machine, precisely

Three kinds of derived material leave your machine, and each is filtered at a different point. Being exact about which filter runs where matters, because they are not the same:

  • Interest signals and the rotation query. Built on your machine from your wiki pages by a construction that can only emit a category from our fixed taxonomy, a vendor slug, or the words of a page's filename. Page content cannot reach this channel by construction, which is a stronger protection than filtering. This channel runs while the program is running.
  • Derived artifacts (the in-conversation task summary, with the category your assistant assigns to it). Written by your assistant inside the session, capped at 500 characters, and filtered when we receive it. No NomosX code runs before your assistant authors this text, so on this path the filter runs at our ingress and not on your machine.
  • Replicated wiki pages. Filtered on your machine before transmission, and before the content is hashed. The graph of vendors that travels with them is filtered on your machine the same way. We do not filter them again on receipt, and we cannot: the hash of the filtered page is the version key the delta sync uses to tell a changed page from an unchanged one, so re-filtering on our side would break replication. This path is client-filtered only, by design.

The filter itself is deterministic pattern matching. It removes:

  • Secrets and credentials: private-key blocks, JWTs, provider API keys and tokens (for example OpenAI, Stripe, GitHub, AWS, Slack, Google), bearer tokens, and assignments of the form api_key: ..., token = ..., password: ....
  • Personal identifiers: email addresses, U.S. Social Security numbers, card-shaped and phone-shaped digit runs, and IP addresses.
  • Machine-local context: file paths, which is where proprietary code lives. Wherever the filter runs, it removes every local path in the text: absolute Unix and macOS paths two or more levels deep, home-relative paths (~/...), Windows drive and network paths, paths with spaces, percent-encoded paths, links that point at a local file (file:// links and editor links), and the paths inside local and development links (such as localhost addresses) and inside links that are not web addresses (such as ssh, sftp and smb links). Only an ordinary public web address keeps its path, and even then any part of it that is itself a local path is removed.
  • Long high-entropy runs: hexadecimal blobs. On the query path, long unbroken base64-style runs are removed as well; on the wiki path that particular rule is deliberately omitted, because ordinary wiki prose legitimately contains long unbroken tokens and the rule destroyed too much legitimate text.

Each removal is replaced by a marker such as [redacted-key].

What the filter does not do, stated plainly. It is pattern matching, not comprehension. It does not detect or remove prose that happens to indicate a sensitive characteristic such as health, religion, political opinion, sexual life or orientation, ethnicity, trade-union membership, or immigration status. Three other controls carry that weight instead, and they are real: sensitive categories are excluded from our matching taxonomy, so an advertiser cannot request them and no such category is ever stored against your profile; we do not attempt to infer sensitive characteristics from anything we hold; and where sensitive content nonetheless reaches us, we delete it on request and do not use it. We are describing the controls we have rather than claiming a filter that understands meaning.

Two further limits apply to everything above:

  • De-linked or deleted after 90 days. Derived artifacts are retained for 90 days for matching quality and billing audit. After 90 days they are de-linked from your account: the account reference is replaced with a constant and location and query hashes are removed, leaving pseudonymous records used only in aggregate. Interest signals are kept as interest counters and rejected-vendor entries on a rolling 90-day expiry: an entry that has not been updated for 90 days is deleted. Sent as an interest signal, the slug of what you are evaluating is stored only as a hash on the request that carried it, and that hash is removed when the request is de-linked after 90 days; the record of each set of signals we receive is de-linked after 90 days in the same way as a derived artifact. They are never sold and never shared with advertisers.
  • Hashed in storage. Query text is hashed before persistence; raw query text is never written to our logs.

If you believe anything we hold captured something it should not have, contact privacy@nomosx.io and we will delete it and review the filter.

Two things that are not derived from your sessions. Every request the Plugin makes names the app it is running in and the Plugin's version, for example claude/0.9.22, and some requests also say how the app displays a sponsored result (whether it draws in a terminal, and whether it shows an interactive card). And while the program is on and you are signed in, the Plugin checks for a new version of itself as you use Claude Code, Codex or Antigravity: at most once a day it reads https://nomosx.io/dist/latest.json, and about every six to seven hours it reads that file again with its signature and, when a newer version is published, may download it ahead of time. Nothing is installed until you run the update command, unless you have set the Plugin to update itself. These checks carry no account, no sign-in token and nothing from your sessions; like any web request, they show your IP address to the services that answer them (Section 5). They do not run after the off command, once your machine has picked up an opt-out of your account (Section 1), or while you are signed out, and setting NOMOSX_NO_UPDATE_CHECK=1 stops them.

For users in the EEA/UK, the legal bases under GDPR are:

PurposeDataLegal basis
Running the program: matching sponsored results to your interest signals and to the current moment, and crediting the impressions you are paid forInterest signals, derived artifacts, impression data, rejected-vendor set, app and versionContract (Art. 6(1)(b)): the agreement you enter is a matched sponsored placement and a 60% share of the value we attribute to it, so the matching is not a way of funding some other service, it is the service itself and the thing your payment is calculated from. Consent (Art. 6(1)(a)) is pleaded in the alternative, withdrawable at any time by the opt-out toggle in your account.
Wiki replication: holding a replica of your local wiki as the matching context you are paid to provide, so you can access and export it, so we can review it under audit when we must, so we can count how many users are active in a category, and so we can compile it into a short structured summary (Section 4a)Replicated wiki pages, intent-edge graph, and the summary compiled from themConsent (Art. 6(1)(a)), given when you install the Plugin and create an account, and withdrawn at any time with the opt-out toggle in your account. Three things make that consent real rather than nominal, and they are set out in full below this table. First, this is one purpose, not a bundle: the thing you are consenting to IS the service, not a separate processing operation attached to some other service you came for. Second, refusing costs you an earning opportunity and nothing else, so the detriment that makes bundled consent involuntary in ad-funded services is absent here. Third, you are paid for it, per impression, on a published share.
Calculating and paying your revenue shareImpression data, account, payout and tax dataContract (Art. 6(1)(b))
Tax, accounting, sanctions complianceTax forms, identity data, payout recordsLegal obligation (Art. 6(1)(c))
Fraud prevention and service securityFraud and security signals, impression dataLegitimate interests (Art. 6(1)(f)): preventing fraud against the exchange, advertisers, and honest users (assessment on file; you may object)
Support, notices, service emailsAccount data, communicationsContract; legitimate interests
Website analytics with cookies on nomosx.io (Google Analytics, Section 9)Website dataConsent (Art. 6(1)(a)), given in the cookie banner and withdrawable from "Cookie settings" in the site footer
Cookieless counts of page views, contact-form submissions and performance on nomosx.io (Vercel Web Analytics and Speed Insights)Website data, with no cookie or identifierLegitimate interests (Art. 6(1)(f)): knowing which pages are read, how many people use the contact form, and that the site loads quickly

Why we think it is fair to ask for this, stated plainly. Every other company in this business takes this data and pays you nothing for it. We take it and pay you 60% of the placement value we attribute to each billable impression, per impression, whether or not you ever click or buy anything. That is the whole deal, and it is the reason the choice in front of you is not the choice those services offer. When an ad-funded service bundles advertising into a product you rely on, refusing costs you the product; that is what makes such consent doubtful. Here, refusing costs you an earning opportunity you did not have before, and nothing else. You keep every other tool you use, unchanged. We think that difference is what makes asking for the whole program, as one thing, a fair request rather than a coerced one. If you disagree, the opt-out is one switch and it is always available.

Withdrawal is as easy as consent, and it is recorded on your account: the opt-out toggle sits in your account settings, next to the rest of your program controls (the same number of clicks as joining), and our servers enforce the result for every machine you sign in from. There is no way to withdraw from part of the program while keeping the rest: one switch stops all of it. The Plugin's off command in Claude Code, Codex or Antigravity, and uninstalling the Plugin, stop it on that machine only and record nothing on your account, so use the toggle if you want the program stopped everywhere. Withdrawal never affects verified earnings, and it does not affect the lawfulness of processing before withdrawal.

Profiling disclosure: while the program is running, matching selects a sponsored result from your interest signals, your rejected-vendor suppressions, and the current derived artifact where the surface provides one. While you are opted out, a machine that has picked up your opt-out (Section 1) sends nothing except, at most once a day, a request for your account's on/off setting, so that switching the program back on in your account reaches it. After the off command, or once you have signed out or uninstalled the Plugin, a machine sends nothing; a signed-out machine also builds no wiki until you sign in again. Commands you run yourself, such as signing in or checking your earnings, still contact NomosX when you run them. This is limited profiling for ad relevance within the Service. We build no audience segment that identifies you and no means for an advertiser to reach you as an individual: the only thing an advertiser sees off your replicated intent graph is a bare count, per matching category, of how many NomosX users are evaluating something in it, with no identity attached (Sections 4a and 5). We make no decision about you producing legal or similarly significant effects by solely automated means, and we never sell your profile or hand it to an advertiser.

4a. Personalization

Personalization is part of the program, not a set of extras. When you install the Plugin and create an account, the program starts, and it has two layers:

  • Baseline personalization: the interest signals in Section 3, used to rank candidate ads, plus rejected-vendor suppression. This layer is what the program pays you for and it is not separable from it.
  • Wiki replication: while the program is on, your local wiki replicates to our servers along with its intent-edge graph.

There is no setting that keeps ads and earning while switching personalization off, and no setting that keeps them while switching wiki replication off. The program runs as one thing: your attention and this processing are what it pays you for. To stop it everywhere, use the opt-out toggle in your account: it stops the whole program, replication included, on every machine. The Plugin's off command in Claude Code, Codex or Antigravity, and uninstalling the Plugin, stop it on that machine only (Section 1).

While the program is on, we process:

  • Interest signals: per-user counters of the matching categories your work touches, with weights, plus your rejected-vendor set and the short slug of what you are currently evaluating. Each expires on a rolling 90-day basis.
  • Replicated wiki: your wiki page text and its intent-edge graph, together with up to the ten most recent prior versions of each page (kept so the profile's development is inspectable and auditable; older versions are pruned automatically). Current and prior versions alike are encrypted at rest (AES-256-GCM); the intent-edge graph is stored in our database with standard access controls. Your data export includes the prior versions we hold, and erasing the replica deletes them with it. Because your local wiki is built by digesting your sessions in Claude Code, Codex (the ChatGPT desktop app included) and Antigravity (Section 2), the replica can reflect activity across those assistants.

What we do with the replica, precisely. Today it serves four purposes and no others: giving you access to it and a copy of it when you ask (Section 8); operator review, which is reason-gated and audit-logged before a single byte is returned; counting, per matching category, how many NomosX users are currently evaluating something in it, which advertisers see as a bare number when they plan a campaign (Section 5); and compiling. Compiling means that when your replica changes, we derive a short structured summary of it (your interests, open intents, vendor relationships and signal weights), kept as a versioned chain so its development over time is inspectable. The compilation is deterministic: it parses the structure your machine already wrote, and no language model on our side ever reads your prose. Each compiled version is encrypted with the same per-user key as the replica, included in your data export, deleted when the replica is erased, and every compile is written to the same access-audit log as a human read. It is not currently read by the system that selects which sponsored result you see: that runs on the interest signals in Section 3. If that changes, we will say so here before it does.

Opting out stops ads, earning, and all personalization use immediately; an opt-out is never reversed as a side effect of any other action, only by you switching the program back on yourself. Opting out does not erase what we already hold at once: you can erase your synced data at any time from your account, and closing your account erases it too. Once you have been opted out for 90 days, we anonymize your data. We keep only statistics that cannot identify you and carry no link to you, and we remove the rest of what we hold that is linked to your account. Text is removed rather than kept, because text cannot be made anonymous: a note about your projects, your tools or the vendors you are weighing can identify you whether or not an account reference is attached to it. A few records stay linked to your account because we need them: your account and its settings, including the opt-out itself; payout, tax and accounting records, so that earnings you have already verified stay payable; the records of your consent, of its withdrawal and of your acceptance of our terms; our log of staff access to your data; and anything we hold for a fraud investigation. Switching the program back on within the 90 days cancels the anonymization; switching it back on later starts afresh, with nothing restored. Copies in our backups expire within 60 days after the anonymization.

Rejected vendors: if you dismiss or reject a vendor, we store that rejection and stop showing you that vendor's sponsored results. We treat this as your standing preference: it continues to apply across sessions, and survives switching the program off and on again, until the entry expires, 90 days after your machine last sent it, until you have been opted out for 90 days, or until you close your account. It is used only to suppress ads, never to target them.

Special categories: we do not seek to collect or infer special-category data, and we never target on it. Sensitive interest categories are excluded from the matching taxonomy, so they cannot be requested by an advertiser or stored against your profile. As Section 3 explains, the deterministic filter removes identifiers and secrets rather than sensitive-topic prose, so incidental special-category content can reach us in free text despite it. Where it does, we delete it on request and do not use it.

5. Who receives data

  • Stripe (payment processing): when you set up payouts, Stripe receives your identity, bank, and tax data and processes it under its own privacy policy (stripe.com/privacy). We receive payout-status metadata. We share your information with Stripe as needed for onboarding, verification, and payouts.
  • Sign-in providers (Google and GitHub): if you choose to sign in with Google or GitHub, that company confirms who you are and shares with us the details listed under account data in Section 2. It handles your sign-in under its own privacy policy and learns that you signed in to NomosX; we send it nothing else about you.
  • Advertisers and affiliate networks: advertisers receive aggregate reporting (billable impressions by category and campaign, weekly; clicks only once click validation is live), never your identity, conversations, artifacts, or wiki. When planning a campaign they can also see, per matching category, a count of how many NomosX users are currently active or evaluating in it. That is a number with no identity attached and no way to reach any individual through it. Where a sponsored result is an affiliate offer, the click routes through the network's tracking (for example CJ or Sovrn) with a click identifier; those networks process click data under their own policies to attribute purchases.
  • Embeddings provider (Google): when an embeddings provider is configured (the Gemini API; Vertex AI in production), the scrubbed query text, or a derived artifact's summary text without its category (after the Section 3 filtering), is sent to Google to compute an embedding on each serve. Google acts as a processor for this purpose under its API data-processing terms; the text is not used by Google to advertise to you. Your replicated wiki pages are not sent to Google.
  • Service providers: cloud hosting, email delivery, analytics, and fraud-prevention vendors, under data-processing agreements, only as needed to run the Service. On nomosx.io these include Vercel, which hosts the site and, with Vercel Web Analytics and Speed Insights, counts page views, contact-form submissions and performance without cookies or any identifier, and which also serves the Plugin's installer and update files from nomosx.io, stored for us with Amazon Web Services; Resend, our email provider, which delivers what you send through the contact form (your email address, role, topic and message) to our team inbox and sends you an acknowledgement; and Google Analytics, which receives website data only if you accept analytics cookies (Section 9).
  • Authorities: where required by law (including tax authorities and sanctions regulators), or to protect rights, safety, and the integrity of the Service.
  • Corporate events: in a merger, acquisition, or asset sale, data may transfer with the business, subject to this policy.

We do not sell your personal information, and we do not share it for cross-context behavioral advertising (as those terms are defined in the California Consumer Privacy Act). Matching is based on your activity within the service you intentionally installed, never on your activity across other businesses' services, and no personal information flows to advertisers. State privacy laws' definitions of "targeted advertising" generally exclude ads based on your current request and on your activity within the requesting service; that is the only kind we serve.

6. International transfers

We are a U.S. company and process data in the United States. For transfers from the EEA/UK/Switzerland we rely on the EU-U.S. Data Privacy Framework (certification pending) and/or the European Commission's Standard Contractual Clauses (2021) with transfer impact assessments, plus the UK Addendum / UK-U.S. Data Bridge for UK transfers. You may request a copy of the applicable safeguards via privacy@nomosx.io.

7. Retention

Rather than silently deleting operational records, we de-link them on a schedule: the reference to your account is replaced with a constant and location and query-hash fields are removed, leaving pseudonymous records used only in aggregate. A worker applies this schedule continuously to every account, including accounts that have opted out.

DataRetention
Derived artifacts (per-request records) and event envelopes90 days, then de-linked from your account (pseudonymous aggregate form)
Click/attribution identifiers13 months, then de-linked the same way
Fraud counters24 months, then removed
Money records (impressions, earnings ledger, payouts, tax)~7 years (tax and accounting law)
Account dataLife of account + 90 days after closure
Personalization: interest counters and rejected-vendor entriesRolling 90-day expiry, then deleted
Personalization: wiki replica and intent-edge graph (including up to 10 prior versions per page)While the program is on, and until you have been opted out for 90 days; you can erase it at any time from your account, and closing your account erases it
Your data, once you have been opted out for 90 daysAnonymized: we keep only statistics that cannot identify you and remove the rest of what we hold that is linked to your account, except your account and its settings, payout, tax and accounting records, your consent and acceptance records, our log of staff access, and anything held for fraud (Section 4a)
Support correspondence24 months
Access requests and contact-form messages on nomosx.io (the email address, role, topic and message)While the request is open, then deleted on request

When you close your account we delete your data or de-link it into pseudonymous form on this schedule, except records we must keep by law (which are retained de-linked from your identity wherever the law allows). Uninstalling removes the Plugin from your assistants and deletes the files the installer added. What the Plugin created while you used it (your sign-in, your consent markers, your local wiki pages and the settings backup) stays on your machine, so a later reinstall does not have to ask you again; run the uninstaller with --purge to delete it too. Uninstalling does not change your account settings (Section 1). Server-side data follows this schedule and can be deleted early by erasing your synced data, closing your account, or exercising your deletion rights.

8. Your rights

Everyone: access, correction, deletion, and a copy of your data; withdraw at any time. Write to privacy@nomosx.io (a request sent to hello@nomosx.io counts from the day we receive it). We respond within the time required by your law (45 days CCPA / 1 month GDPR, extendable as permitted). If we deny a request you may appeal by replying to our decision; we will have a different reviewer assess it. You will not be discriminated against for exercising rights (see the financial-incentive note below).

EEA/UK/Switzerland: rights of access, rectification, erasure, restriction, portability, objection (including to legitimate-interest processing), and withdrawal of consent; and the right to lodge a complaint with your supervisory authority (for example, your national DPA, or the ICO in the UK).

California (CCPA/CPRA): rights to know, delete, correct, and port; to limit use of sensitive personal information (we do not use SPI to infer characteristics, so the limit right's trigger does not apply); and to non-discrimination. We do not sell or share personal information, so there is no sale/share opt-out to exercise; we nonetheless honor Global Privacy Control signals on nomosx.io as a refusal of analytics cookies (Section 9), and in-product settings govern the Plugin. Authorized agents may submit requests with proof of authorization. Categories collected are listed in Section 2; retention in Section 7.

Other U.S. states (Colorado, Connecticut, Virginia, Utah, Texas, Oregon, Montana, Maryland, and others): equivalent rights of access, deletion, correction, and portability, and the right to opt out of targeted advertising and sales (we do neither, as defined by those laws: our ads respond to your activity within the Service you installed). Maryland residents: we do not sell sensitive data and we collect only what is reasonably necessary for the Service you requested.

Notice of Financial Incentive (California)

The NomosX earnings program pays you for participating in a service that processes the data described in this policy, so California treats it as a financial-incentive program:

  • Material terms: you receive 60% of the placement value NomosX attributes to each verified billable impression served in your sessions, up to 50 sponsored results per day, paid via Stripe per the Terms of Service. The program processes the interest signals and derived artifacts described in Section 3; that processing is what the incentive compensates.
  • Opt-in: installing the Plugin and creating an account, which starts the program. Withdrawal: the opt-out toggle in your account, at any time, without penalty; verified earnings remain payable. Earning and the processing it compensates are the same act: there is no way to earn without the processing described in Sections 3 and 4a, wiki replication included. That is the bargain, and it is the reason the incentive exists: we are buying access to context that other advertising businesses take for nothing.
  • Value of your data: we estimate the value of the data you provide as equal to the payments we make to you: 60% of the placement value NomosX attributes to each impression. Method: NomosX sets a per-impression rate for each earning surface from what advertisers pay for comparable placements across the exchange, and the program passes through a fixed majority share of that rate on every billable impression, so the incentive is directly and reasonably related to the data's value to us (the remaining share funds the exchange).

9. Cookies and local storage

Plugin (your machine): the Plugin stores an authentication token, a per-install OAuth client credential, your settings, and its local processing data (including your local wiki pages). The token and per-install credential are strictly necessary to deliver the Service you signed up for (serving, capping, and crediting impressions to your account); local processing data never leaves the machine except as described in Sections 3 and 4a. Lifetimes: the sign-in token lasts for your session and renews while you stay signed in; the per-install credential lasts for the life of the install. Installing and logging in constitutes your consent to this storage. Uninstalling removes the Plugin from your assistants and deletes the files the installer added. What the Plugin created while you used it (your sign-in, your consent markers, your local wiki pages and the settings backup) stays on your machine, so a later reinstall does not have to ask you again; run the uninstaller with --purge to delete it too. Uninstalling does not change your account settings (Section 1).

Sign-in (auth.nomosx.io and console.nomosx.io): when you sign in, our sign-in service at auth.nomosx.io sets a session cookie named __Secure-better-auth.session_token, which lasts 7 days and is renewed as you use your account, plus one more whose name begins the same way for each account signed in on that browser, which lasts 7 days from the sign-in that set it. During a sign-in step, such as a Google or GitHub sign-in or a passkey, the sign-in service also sets short-lived cookies that last ten minutes or less. The console at console.nomosx.io sets its own session cookie, nomosx_console, which lasts up to one hour and is renewed while the console is open. Each of these cookies is sent only to the site that set it. When you sign out, the sign-in service deletes its session cookies. The console deletes nomosx_console when you sign out in the console or the next time the console opens and finds you signed out; otherwise the cookie expires on its own within the hour. The session and sign-in step cookies are strictly necessary to keep you signed in, and we do not use them for analytics or advertising. The console's side menu remembers whether you left it open or closed in a cookie named sidebar_state, kept for 7 days, and the console keeps your light or dark theme and the console view you last used in your browser's local storage. If you sign up with Google or GitHub, the sign-in service keeps the confirmations you gave on the sign-up form in your browser's local storage until you come back from Google or GitHub, so that it can record them against your new account; it deletes them then, and it does not use them after 15 minutes. These are not used for analytics or advertising either.

Website (nomosx.io): the site sets no cookies unless you accept analytics. If you accept in the cookie banner, Google Analytics sets cookies (one named _ga and one named _ga_ followed by a code for our site, each kept for up to two years) to count visits and see how people use the site. Nothing from Google loads before you accept, and Google's advertising features (Google signals and ad personalization) stay off. If you decline, or your browser sends a Global Privacy Control signal, Google Analytics does not load and no analytics cookie is set; you can change your choice at any time from "Cookie settings" in the site footer, and declining there stops Google Analytics at once and deletes its cookies. Vercel Web Analytics and Speed Insights count page views, contact-form submissions and performance without cookies or any identifier. If you choose a light or dark theme or answer the cookie banner, the site stores that choice in your browser's local storage.

10. Security

We use encryption in transit and at rest, access controls, secrets management, and vendor due diligence. Replicated wiki page text is encrypted at rest with AES-256-GCM under a per-user key, and operator access to it is reason-gated and written to an access audit log before any content is returned. No system is perfectly secure; we will notify affected users and regulators of data breaches as required by applicable law, including within 72 hours to EU supervisory authorities where GDPR applies.

11. Minors

The Service is for adults. We do not direct the Service to, and do not knowingly collect personal information from, anyone under 18. If we learn we have collected personal information from a person under 18, we will delete it and close the account. Contact privacy@nomosx.io if you believe a minor has provided us data.

12. AI transparency

Every sponsored result is labeled as sponsored. That is our disclosure obligation and we meet it on every placement, on every surface, in every format. The label reads "Sponsored", and every result also carries a machine-readable sponsorship marker.

Who made the advertisement, and who says whether AI made it. Advertisements are ordinarily written by advertisers, not by us. We select which advertisement to show you and when; we do not write it. Whether a given advertisement was produced with AI is therefore something its author knows and we do not, so the obligation to disclose that sits with the party who created it: the advertiser is responsible for any further disclosure its creative requires, including whether it was made with AI.

One exception, stated because it is ours. For some affiliate offers NomosX writes the one-line pitch itself rather than receiving creative from the merchant. For those pitches the disclosure obligation is ours, not the merchant's, just as it is the advertiser's for creative an advertiser supplies.

The matching system. It selects advertisements; it does not generate them. It performs no emotion recognition and no biometric categorization, and it makes no decision about you in any high-risk category under the EU AI Act.

13. Changes to this policy

We will post changes here and notify you by email or in-product for material changes, at least 14 days before they take effect. If a change expands how we process data already collected, we will ask for consent where the law requires it. In particular, if we begin using your replicated wiki to select which sponsored results you see, we will say so here first.

14. Contact

Nomos Arc, Inc., Attn: Privacy, privacy@nomosx.io

Version tos+pp-v12-2026-11-10