Privacy Policy

The short version

We built Axon NeuroAutomata for researchers who care about their data. Here's what that means in practice:

• Your protein sequences are not stored. They're processed in memory, held in a temporary queue for up to one hour so you can grab your results, and then they're gone. We don't have a database of your sequences. We can't look at them later even if we wanted to.
• We don't train AI models on your data. Your sequences are never used to train, fine-tune, or improve ESM-2 or any other model. Period.
• We know very little about you. Your email address (for login), your subscription tier, and anonymous usage analytics. That's it.

What we collect and why

Your email and account info

When you sign up, Clerk (our authentication provider) collects your email address and manages your login session. We never see or store your password — Clerk handles all of that.

Why: So you can log in and we can tell free accounts from Pro accounts.

Your protein sequences (temporarily)

When you submit a job, your sequences sit in Redis (an in-memory data store) for up to one hour. During that time, they're processed on GPU infrastructure, and the results are made available for you to view and download. After one hour, both the sequences and results are automatically deleted.

Why: To run the analysis you requested.

For protein language model analysis (ESM-2, ESMC, and SaProt — embeddings and mutation scoring), your sequences are sent to Modal (modal.com), a serverless GPU provider. Modal processes them in isolated containers and deletes inputs/outputs once results are retrieved (max 7 days). Modal never accesses your data.

Structure prediction (a Pro feature) is the exception: it is performed by an external third-party service. Your sequence is sent to the ESMFold public API (api.esmatlas.com), which is operated outside Axon — we have no data-processing agreement with it and don't control how long it keeps submitted sequences. Details in "Third-Party Service Providers" below.

You can also submit 3D structure files (PDB or mmCIF format) for structure-aware analysis — either uploaded directly, fetched from the RCSB Protein Data Bank, or predicted by ESMFold. Structure files are held in the same temporary queue as your sequences (up to one hour) and are sent to Modal, which encodes them into a compact structural representation (via Foldseek) for the model to use.

Usage analytics

We use PostHog to understand how people use the platform — which features get used, how long jobs take, where people get stuck. We track things like "a user submitted 5 sequences for embedding generation" — never the sequences themselves.

What PostHog sees: Page views, feature usage events, your email (for user identification), subscription tier, browser type, job counts and durations.

What PostHog never sees: Your protein sequences, embeddings, mutation scores, or any analysis results.

We configure PostHog with:

• No session recording (we don't watch your screen)
• Nothing tracked until you make a cookie choice — decline and PostHog runs cookieless (no localStorage, no cookie); consent and it stores an anonymous ID in localStorage plus a cookie
• Click tracking on interactive elements only (links, buttons, form controls) — never the content you type
• No pageview auto-capture; disabled on localhost during development

Why: To improve the product and understand which features matter.

Operational traces

We use Logfire (by Pydantic) for distributed tracing — it helps us debug performance issues and find errors. As of 2026-06-30, traces are configured to exclude the contents of your submissions: your protein sequences, any structure files you upload and the structural tokens (3Di) we derive from them, and the results we derive from them (embeddings, mutation scores, predicted structures) are no longer sent to Logfire. Traces recorded before that date may still contain submitted sequences and age out under Logfire's retention window.

What Logfire sees: job timing data, error details, and HTTP calls to external services (such as ESMFold and PDB). What Logfire no longer receives (since 2026-06-30): your protein sequences, uploaded structure files and the 3Di structural tokens derived from them, or the results computed from them.

Why: So we can debug when things break and keep the service running well.

Payment information

We never see your credit card number. Billing is handled by Clerk, which uses Stripe as its payment processor. All we know is whether you're on the Free or Pro plan. Card data, billing address, and payment details are between you, Clerk, and Stripe.

Why: So Pro features work for Pro subscribers.

Security and server logs

Our servers keep access logs — including your IP address, browser user-agent, and basic request details like the page requested and the time. These logs leave out your login cookies and authorization headers, are deleted within 30 days, and are never used for analytics or advertising, or sold or shared.

Why: To keep the service secure and reliable and to prevent abuse.

Cookies

We use a small number of cookies:

That's it. No third-party ad trackers, no social media pixels, no fingerprinting.

Who else gets your data

We don't sell your data to anyone. We don't share it with advertisers. We don't monetize it in any way beyond providing you the service you signed up for.

Your rights

Delete your account: You can delete your account through Clerk's account settings. This removes your login, email, and subscription data.

Delete your analytics data: Contact us at privacy@axonagentic.ai and we'll submit a deletion request to PostHog on your behalf.

Export your data: Contact us and we'll provide what we have — which, honestly, isn't much given our ephemeral architecture.

Opt out of analytics: Use the cookie consent banner on your first visit to decline analytics. You can also contact us at privacy@axonagentic.ai to opt out or request deletion of your PostHog data.

Your sequences and structure files: They auto-delete after one hour. There's nothing to request deletion of — it's already gone.

If you're in the EU/EEA, you have rights under GDPR (access, rectification, erasure, restriction, portability, objection). If you're in California, you have rights under CCPA (know, delete, correct, opt-out). We honor all of them. Email privacy@axonagentic.ai and we'll respond within 30 days.

Data retention at a glance

AI and your data

Axon uses several protein language models — ESM-2 (Meta AI Research), ESMC (EvolutionaryScale / CZ Biohub), and SaProt (Westlake University; sequence-only 1.3B and structure-aware 650M variants) — to generate embeddings, mutation scores, and other outputs. These are computational predictions — not experimental results, not clinical diagnoses.

We do not use your sequences or structures to train any AI model. Model weights are loaded once and used for inference only. Your data goes in, results come out, and your sequences are deleted. Nothing about your submissions influences the models.

Children

Axon NeuroAutomata is not intended for users under 16 (EU/EEA) or 13 (US). If we discover we've collected data from a minor, we'll delete it promptly.

Changes to this policy

If we make material changes, we'll notify you by email at least 14 days before they take effect. We'll also update the "Last Updated" date at the top.

Contact us

For privacy questions, data requests, or concerns:

Email: privacy@axonagentic.ai
Operator: Jonathan Agoot, Axon Agentic

EU/EEA users: You have the right to lodge a complaint with your local supervisory authority.
UK users: Contact the Information Commissioner's Office (ICO).
Canada: Contact the Office of the Privacy Commissioner of Canada.