Control & privacy
If you let an AI work on your data, you should be able to see exactly what it reached and undo what it did. Nolocron keeps your data on your own disk, lets you scope what any assistant can see, records every action it takes, and keeps what the AI said separate from what your sources said. The control is real because the data never leaves your custody to begin with.
Your sources, your switches
Receipts
Imports, exports, deletes, scheduled tasks, assistant queries — an append-only record captures what happened, so any operation can be reviewed after the fact. Every assistant action leaves a receipt of what it accessed, and answers cite the items they drew from. The record-keeping runs today; a single view that shows you everything an AI saw and did, with one-click undo, is on our roadmap — we describe it as direction, not a finished screen.
The trail behind one answer
SCOPE
Held to project ‘Q3 Launch’ (locked)
READ
12 blocks from ‘Launch notes’
TOOL
vault.search — query: ‘pricing tiers’
TOOL
block.hydrate — 3 blocks expanded
CITE
4 sources cited in the answer
RESULT
Answer returneddone
RUN RECEIPT
Receipt saved · 7f3a91e2-04bd
Recorded in the append-only event store — reviewable after the fact.
The guarantees
Nolocron tracks the actor behind every action — you, an assistant, an importer — each with an explicit, scoped identity. Access is default-deny and least-privilege; scope can tighten but never silently widen mid-task, and even you act through a scoped identity. There's no master key that quietly sees everything.
Imported source material is canonical and never overwritten. Machine-generated additions — summaries, embeddings, classifications — are labelled derived and kept distinct. A model's guess can never quietly become the record, and the canonical copy is always recoverable underneath it.
Why it matters
For legal, medical, and financial work, "the AI looked at something" isn't good enough — you need to know exactly what, under whose authority, with a record you can hand to someone else. Grant an assistant exactly one client, project, or source; when it does real work it happens in a sandbox that can never corrupt your data, fully reversible. The receipt and the undo are what turn scoped AI access into something you can defend.
Scoped to one client — with a receipt
ASSISTANT RUN · RECEIPT
Recorded in the append-only event store — a trail you could hand to compliance.
Custody
Control is only as real as the data underneath it. Your archive lives on your own disk, not in a required Nolocron cloud account; it's stored in standard, inspectable formats (SQLite, content-addressed files, Markdown and CSV exports) you can back up, move, copy, or delete like any other files; external apps connect through an open protocol (MCP) under the same scope limits — nothing gets a side door. Because it's local, the controls above are enforced by construction, not by a policy a vendor could quietly change.
Preview limits
On your terms
Local-first, scoped, and receipted — so trust is something you verify, not something you extend.
