Your company's knowledge stays on hardware you own.
Most AI tools require you to send data to their cloud. ADAMAS does not: the vault, the ledger, and the AI run on a machine you own. This page sets out what stays local, what is transmitted, and which third parties are involved.
The local-first architecture
An ADAMAS deployment is a local vault — a structured decision ledger stored on a machine in your office (typically a Mac Mini M4 you purchase and own) — plus Hermes, the AI agent that reads your sources, extracts decisions, and generates assets. Hermes runs on that same machine, 24/7.
There is no ADAMAS cloud account, no central database of client vaults, and no telemetry from your vault back to us. If you unplug the machine, your entire system — data and AI included — is offline and intact.
What stays local — and what doesn't
Never leaves your machine
- The decision ledger itself — every decision, context, and trade-off
- Source material ingested from your tools (email, Slack, CRM, docs)
- Generated assets (onboarding docs, frameworks, investor materials)
- Embeddings, indexes, and the AI's working memory
Transmitted — only with your opt-in
- Hybrid-cloud tasks you explicitly approve (see flow below)
- Software updates pulled to your machine (nothing pushed out)
- During the build engagement: material you choose to share with us for setup
The hybrid-cloud opt-in flow
Roughly 80% of daily workloads run fully locally. For rare heavy tasks where a frontier cloud model is genuinely better, ADAMAS supports an optional hybrid route — and it is opt-in per task, never a default:
ADAMAS flags that a task would benefit from a cloud model and shows you exactly which content would be transmitted.
Nothing is sent until you explicitly approve that task. Declining means the task runs locally instead (slower, but private).
Approved content goes to the cloud provider for that single task; the result is stored back in your local vault. The ledger records that the task used the hybrid route.
Data ownership
Your vault is your property — outright. The hardware is purchased and owned by you. The ledger data is yours. The generated assets are yours. If our engagement ends, everything keeps running and nothing is withheld: the ledger is stored in open, portable formats (structured Markdown + JSON), so you can read, export, or migrate it without us.
We keep no copy of your vault. We could not hand your data to a third party even if asked to — we don't have it.
GDPR posture
The local-first design is the GDPR answer: your company data is processed on premises, under your control, in your jurisdiction. For client engagements, Falcon Intelligence Group acts as a processor only for the material you share with us during the build — governed by a data processing agreement (DPA) — the template is public.
This website itself sets no tracking cookies. Analytics run on Plausible, a cookieless, EU-hosted, privacy-first service. Details are in our Privacy Policy.
Third parties & sub-processors
ADAMAS is built on third-party components. Here is each one and what it touches.
| Component | Role | Where your data goes |
|---|---|---|
| Hermes (third-party AI agent) | The AI engine that powers ADAMAS. Property of its respective owner. | Runs locally on your hardware. Your vault data is processed on your machine, not on Hermes' servers. |
| Cloud AI provider (hybrid route) | Optional heavy-task processing | Only the content of tasks you explicitly approve, per task. |
| Netlify | Hosting for this website and its contact forms | Form submissions you send through this site. |
| Plausible Analytics | Cookieless website analytics (EU-hosted) | Anonymous, aggregated page-view data from this site only. |
| Google (Calendar booking, Fonts) | Call scheduling; one webfont on this site | Booking details you enter; standard font-delivery requests. |
Found a vulnerability?
Write to security@adamas-project.com. We read every report and reply within one business day.
Want the architecture walked through against your own setup?
Book a Clarity Audit →