Privacy Policy

Axiom is a developer platform for publishing, composing, and invoking software services. It is operated by Christian George Lucas, an individual — there is no company behind it today, and so no privacy department. For anything in this document — a question, a correction, a deletion request — email christiangeorgelucas@gmail.com. That address is the whole privacy contact; there is no separate DPO, and we make no certification claims (no SOC 2, no ISO 27001, no HIPAA).

• Account data. If you sign up with email and password, we store your email address and a hash of your password (Argon2id — we never store the password itself). If you sign in with GitHub or Google, we store the profile those providers return: your email, display name, avatar URL, and their account identifier. We also store the handle you publish under.
• Your content. The packages, nodes, flows, and source code you push or publish, plus their metadata. Published artifacts are public by design — that is what a marketplace is.
• Execution data. When you invoke a node or a flow, we process the input you send, the output it produces, and the execution logs and traces needed to run it, show you what happened, and debug failures.
• Secrets. Credentials you register on the Secrets page, so your flows can authenticate to third-party services. See “How secrets are handled” below.
• Builder conversations. If you use the AI app builder: the prompts you give it, the specification it derives from them, the code it writes, and the files it reads while working. This is what building an app with an AI agent consists of — see the Anthropic entry below for where it goes.
• Product telemetry. How the product is actually used: which commands and console actions you run, what the AI builder tried, which capability searches found nothing, which builds failed and why, error codes, retries, and timings. This is how we find out what is missing and what to fix.
• Operational logs. Request metadata — IP address, user agent, timestamps, paths, response codes — used to run the service, enforce rate limits, and investigate abuse and outages.
• Product analytics on the marketing site. The public site (landing, marketplace, docs) uses PostHog to count page views and interactions so we can tell which parts of the product people actually reach.

Axiom gets better by watching itself get used. We look at what you build here and how it runs — the shape of your flows, what the builder searched for and could not find, what failed, what you had to write by hand — and we use that to fix the platform and to decide which reusable components are worth building next. When we build one, we write it ourselves and publish it under our own name, so the next person's app is easier to make than yours was. What we learn that way — the patterns, structures, techniques, and gaps — and anything we build from it belongs to us; your artifact itself stays yours and stays out of the marketplace. We may also use content stored here to develop and improve the AI models that power the product.

Where our own training happens. We do not currently send your content to any third party in order to train our models. If that changes, we will name the provider here before it starts.

You can turn off AI model training whenever you like. It is one switch in Console → Account. It is free, we never ask why, and nothing else about your account changes. It applies from the moment you flip it — we cannot pull content back out of a model that is already trained, and we would rather say so than imply otherwise. What the switch covers, precisely: it stops us using your content to train models. It does not stop us running the service, and it does not stop us using what we learn to improve Axiom and to decide what to build next.

• We do not sell or rent your data to anyone.
• We do not run advertising, and we do not share data with ad networks.
• We do not use the data your application collects from its own end users for anything except running your application. Your users' records are not our material.
• We do not read the secrets you register. They are encrypted, never displayed back, and released only to an execution you have granted them to.
• We do not publish, list, sell, or make your private packages, flows, or applications available to other users, and we will never mark one of them as a candidate for the marketplace. It stays yours, and it stays out of the marketplace.
• We do not ask for payment details, so we never hold card numbers or billing information.

A secret you register is encrypted at rest and is never shown back to you or returned by any page or API response — the Secrets list displays only the name and the date added. Decryption happens only when an execution that is entitled to that secret starts, and a node receives a secret only when you have explicitly granted it. Secrets are scoped to your tenant: another user's node cannot read them, and they never appear in package source or in a flow definition. The mechanics are documented in Manage secrets in a flow.

A secret you hand to a node is, by construction, usable by that node's code. Grant secrets only to nodes whose source you have read — every published node's source is readable before you invoke it.

Axiom runs on third-party infrastructure. The subprocessors we use today:

• Google Cloud Platform — hosting, compute, managed Postgres, and container registry. All service data lives here.
• Cloudflare — DNS and edge protection for our domains.
• PostHog — product analytics on the public marketing site.
• OpenAI — two things. Catalog enrichment: when you publish a package, node, message type, or flow to the public marketplace, we send its listing text and its handler source to OpenAI to write the catalog description and tags and to compute the search embedding. On this path, an artifact you have only pushed and not published is never sent. Memory: text a node writes to, or looks up in, Axiom Memory is embedded so that semantic recall works — that path is about your memory contents, whether or not anything is published. Your execution inputs and outputs, your logs and traces, and your secrets are not sent to OpenAI.
• Anthropic — two things. Docs assistant: the “Ask AI” assistant on the docs site sends your question and the documentation to answer it. The AI app builder: the builder is built on Anthropic's Claude models, so your prompts, the specification it derives from them, the files it reads in your workspace, and the code it writes are sent to Anthropic to generate your app. There is no way to use the builder without this. We use Anthropic's commercial API. As of August 2026 its terms provide that Anthropic may not train its models on customer content from those services, so this content is processed to answer the request and is not used to train Anthropic's models. That is their commitment, not ours to make — check their current terms if it matters to you. The AI model training setting described in the Terms of Service governs our models, not theirs.
• YouTube — the demo video on the landing page is embedded through youtube-nocookie.com, which does not set tracking cookies until you press play.

If you sign in with GitHub or Google, that provider necessarily learns that you authenticated with Axiom. These providers operate in the United States and elsewhere, so using Axiom involves transferring your data internationally. This list will change as the product does; the effective date above tells you how current it is.

Axiom is in public beta and is explicitly not for production use. As the banner on every page says, data may be wiped at any time — beta environments get reset, and we do not promise that anything you store survives. Do not keep the only copy of anything here.

You can delete your own content from the console: secrets, API keys, pushed packages, and published versions all have a delete action, and deletion takes effect immediately for new executions. To have your account and its remaining data removed, email the address above and we will delete it. Operational logs age out on their own retention cycle and are not kept indefinitely.

Two exceptions worth stating plainly. Once you publish a package or flow to the public marketplace, other people may have already installed or depended on it, and copies may persist outside our control. And copies and derivatives already created before you deleted — AI models already trained, aggregated data already derived, and components we have already built — survive deletion of your content and termination of your account. We cannot un-train a model or un-publish a component we already wrote. Deleting content stops us using that content going forward; it does not reach back into what was already made from it.

Traffic is served over TLS. Passwords are hashed with Argon2id and checked against known breach corpora at signup. API keys are stored only as SHA-256 hashes, expire on inactivity, and are revocable from the console. Every request is scoped to your tenant by the API gateway, which stamps your identity server-side — a client cannot ask for another tenant's data by supplying a different header. No system is perfect; if you find a security problem, email the address above and we will fix it and tell you what happened.

Axiom is for adults. You must be at least 18 to hold an account, the service is not directed at children, and we do not knowingly collect their data. If you believe a child has an account, email us and we will remove it.

You can export or delete your content from the console at any time, and you can ask us for a copy of the account data we hold or for its deletion by email. AI model training is a single switch in Console → Account that you can flip yourself, either way, without asking us and without giving a reason. If you would rather not be counted by analytics on the marketing site, a browser-level tracking blocker is effective — the site works fine without it.

When this policy changes materially we will update the effective date and note the change on the site. Continuing to use Axiom after a change means the updated policy applies to you. See also the Terms of Service.