Privacy
Privacy is architecture here, not a policy bolted on. Most of our software runs on your machine, and we designed the parts we do host to need as little of your data as possible.
Your tool calls, policies, and memories stay on your device. Our servers store only the minimum metadata needed to run your account — never sold, never used to train models, never read out of curiosity.
The website you’re reading
No analytics. No trackers. No cookies. This site is static HTML, and it sets none of these. The fonts are self-hosted, so simply reading the site makes no third-party requests at all, and nothing you browse is ever submitted to us.
Our forms. The contact form does send us what you type — that’s how we reach you back. Nothing leaves your browser until you press the button. We use what you send only to answer you: it isn’t used for marketing, and it is never shared or sold. Delivery runs through Resend, our email provider, who process the message on our behalf in order to send it.
The site is served from Cloudflare’s edge. Their infrastructure may keep standard server logs as part of serving traffic (see Cloudflare’s own privacy policy for what they record); we add nothing on top of that.
Guardrail — the engine
The guardrail engine is self-hosted software that runs on your infrastructure. From the commercial license, §2:
The engine itself is self-hosted software on your infrastructure. It does not phone home to enforce billing.
In practice:
Your policies, your audit logs, and your check data all stay on your hardware.
The hosted console
The one service we run for you is the Guardrail Console at console.ellmstack.dev — organizations, members, API keys, billing, and usage tracking. From the commercial license, §15:
the Hosted Service processes the minimum data needed to run the service (account, organization, member, key metadata, and guardrail event records your engine instances report). We do not sell your data and do not read your event payloads except to operate the service, debug issues you report, or comply with law.
| What we store | What we do with it |
|---|---|
| Account & organization metadata | Run the service — who belongs to your org, what plan you’re on |
| API key metadata | Issue, rotate, and revoke keys; enforce your plan’s key limits |
| Event records your engines report | Measure check volume, surface usage, send the 80%/100% alerts you signed up for |
| Event payloads | Read only to operate the service, debug issues you report, or comply with law — never sold |
Payments are processed by Stripe. Checkout happens on Stripe’s pages; we don’t see or store card numbers — we receive only subscription and payment status events.
What we never do: sell your data, use it to train models, or read your payloads out of curiosity. There’s no ad business here; the products are the revenue.
Engram — the memory vault
Engram is a local-first, encrypted memory vault. Your agent’s memories are stored in SQLCipher with keys bound to your hardware, on your machine. Capture, consolidation, decay, and retrieval all run locally. The only thing that ever leaves your machine is the assembled context you choose to send when you call your model provider — the same as any prompt you write.
Amparo — the agent that acts under policy
Amparo brings its own LLM: model calls go to the endpoint you supply, never through us. The agent is local-first, and its privacy ledger records counts, never values — what it did, not the data it saw. The hosted operator surface at amparo.ellmstack.dev keeps only what it needs to run the app: your session state and approval records.
Questions
If you have questions about how we handle data, or want to report something, email Elai-intelligence@pm.me. The full commercial terms — including §15, quoted above — are in the commercial license (request a copy).