One diagram. One hard boundary.
This is the entire ADAMAS data flow. The dashed gold line marks your hardware: ingestion, evaluation, the ledger, and asset generation all run inside a machine you own. The only thing that ever crosses it is a task you explicitly approve.
Fig. 1 — The ADAMAS data flow. The dashed gold boundary is physical: a machine in your office that you own.
The five stages
Sources. Your decisions already happen in email, Slack, your CRM, meeting transcripts, and docs. Nothing about how your team works has to change.
Ingestion. Connectors pull from those sources onto your machine. Data flows in, toward your hardware — never out to a vendor cloud.
Evaluation. Hermes — the third-party AI agent that powers ADAMAS — runs locally, 24/7. It separates noise from decisions and extracts the why, the who, and the trade-offs.
The ledger. Every decision is stored structured and bi-directionally linked, in open formats (Markdown + JSON) you can read and export without us. Click through a sample ledger →
Asset generation. Onboarding docs, hiring frameworks, investor materials — generated from the ledger, regenerated when decisions change. Read a generated sample →
The boundary, precisely
Inside the dashed line: source material, the ledger, embeddings and indexes, the AI's working memory, and every generated asset. None of it is transmitted anywhere. There is no ADAMAS cloud account and no telemetry from your vault to us.
The single exception is the hybrid route: for rare heavy tasks, ADAMAS can ask your permission to send one task's content to a frontier cloud model. You see exactly what would be sent, you approve or decline per task, and the result is stored back in your vault. Details on the Security page.
Want this mapped onto your actual tool stack?
Book a Clarity Audit →