Everything you can do —
by who you are and what you need.

An API key in two fields, then the CLI or MCP.

1. Fill "Work email" and "Company", accept, click "Get my API key →".
2. Your key is shown once — "Copy" it and tick "I saved my API key".
3. "Set up my 12 words →" — the recovery words for your organization. The console opens only after both.
4. Install the CLI and learn the concepts from the terminal.

npm i -g @nodatachat/nodata
nodata login --email you@company.com
nodata explain            # governed-transaction, capsule, grant, revoke-vs-burn …
nodata scan ./my-folder    # what is sensitive here?

Open a tenant, secure recovery, then install nothing — or the right-click sealer for Windows.

1. Fill company, work email and role, then "Activate · start free trial". The sign-in link arrives by email.
2. Set up recovery right away (see "What if I lose the key?").
3. Console › Install: "Scan now" (a folder, in the browser), "Open Seal Studio", "Open verifier" — no install needed.
4. Optional: Windows right-click sealing — download from /seal-download, then pair the computer in Console › Seal devices ("Connect this computer").

One consent screen; then scan, recommend and enforce on the real files.

1. Microsoft: "Connect Microsoft 365 now" and approve as a Microsoft admin. The page lists every scope requested.
2. Google: "Connect Google Workspace" and approve.
3. Connectors shows the full 6-step activation and what is still missing.

Two independent roots: 12 words, and a password-protected recovery file. Set both on day one.

1. "Create master key" → write down the 12 words (copy / download / print) → "I saved them — continue".
2. Choose a password and "Download recovery file (.ndc.html)".
3. Prove it works: upload the file, enter the password, "Open the file and verify".
4. Confirm with the 6-digit email code, or biometrics as a third root.

Deny by default. Give the AI its own identity and allow exact columns and classifications; for everything else the decryption key is never even derived.

1. Console › AI agents › "+ New agent".
2. Identity: a handle (lowercase, hyphens) and what the agent does.
3. Scope: tick the allowed columns (e.g. first_name, but not id_number), filter by team, and choose allowed classifications (public, internal, restricted, attorney_client).
4. Validity: 30 / 90 / 180 days or a year.
5. "Apply · register agent". The token is shown once — give it to the agent, never to a person.
6. Anything outside scope is refused (403) and appears under "Blocked attempts · last 7 days", each with a verifiable receipt.

Admin mode registers the table and grants; agent mode can only use the grant it was given.

1. Admin mode (your org key): register the table and name which columns are encrypted.
2. Grant: agent + table + columns + optional WHERE + classifications + TTL. You get an ndca- grant token.
3. Agent mode: start the agent with only that grant token. Its loop is decide → retrieve → read.
4. A column outside the grant returns 403 scope_violation with denied_columns — the model never sees it.

# admin: register + grant
claude mcp add nodata -- npx @nodatachat/mcp --api-key ndp_YOUR_KEY
#   tools: nodata_register_table · nodata_grant · nodata_revoke

# the agent itself: only its grant
claude mcp add nodata -- npx @nodatachat/mcp --grant-token ndca-YOUR_GRANT
#   tools: nodata_decide → nodata_retrieve → nodata_read

Scan the table, review the proposed AI exposure policy, approve it.

1. Run a scan on the table — the scan proposes a policy.
2. Review each proposal: "AI will see", "Blind to" and blocked classifications. It is marked "Proposed".
3. "Approve & activate". The model sees only the allowed columns; the previous active policy is archived, not deleted.

A read-only simulation over your scanned folders.

1. Pick a scanned root under "Scanned roots".
2. Choose the classes you are thinking of granting under "Simulated grant".
3. Read the three tiles: allowed to AI · blocked (above level) · blocked (not scanned).

Freeze it. Every credential it holds stops working on its next request.

1. Web: in the Courier hub, find the agent row and press "freeze", then confirm.
2. Terminal: nodata agent revoke <handle>.
3. MCP (admin mode): nodata_revoke with the handle and a reason.
4. If the agent belongs to an employee, disabling the employee also cuts the agent.

nodata agent revoke crm-assistant
nodata explain revoke-vs-burn

Yes — the Gateway keeps your provider keys encrypted and records evidence without storing prompts.

1. Enter your NoData API key, add a provider key (OpenAI, Anthropic, Google, Mistral…), "Encrypt & Save".
2. Point your SDK’s base URL at the gateway. Your code never holds the provider key again.
3. "Revoke" removes a provider key instantly.

In the policy composer: a default envelope, plus rules that fire on conditions. Simulate, then enforce.

1. Console › Organisation rules › "Open the policy composer →".
2. Set the default envelope — what applies when no rule fires.
3. "+ New rule": a reason (it appears in every receipt and audit), then conditions — all must hold. Available: file family, folder path, scan findings, classification, destination channel, external destination, sender, sender department, sender role, amount, send hour.
4. Outcome: "Set" (exactly this envelope) or "Floor" (only ever raises protection). Choose the envelope and a priority.
5. "Simulate" — see which rules fire on your real data before anything changes.
6. "Save & enforce".

A folder policy: allow-list, lifetime, open count, burn, and which options employees see when they right-click "Seal".

1. "➕ New policy". Folder name or label — use * for the whole organization.
2. "Who may open": an email allow-list. When filled, it overrides the recipients an employee types.
3. Lifetime (1 hour → unlimited) and "max opens".
4. What employees see in the Seal window: no time limit · 30 minutes · 2 hours · burns after 8 hours · one open per recipient — and which is the default.
5. Burn on first read, and the delivery channel: view only · watermarked download · capsule only.
6. "Save policy" — a receipt is created.

The access ladder: each rung asks for more proof and unlocks higher classifications.

1. Turn on "Ladder enabled".
2. "+ Add secret" (a word, code or PIN). It is shown once — "I saved them — hide".
3. Build rungs: each rung "requires all of these" and unlocks the next classification.

Let the scan tell you: risk clusters become one-click draft policies.

1. No scan yet? "⚡ Run scan", or "scan-local in browser" — nothing leaves your machine.
2. Each risk cluster card opens a new folder policy, pre-filled.
3. Review, adjust, save.

Dry-run first — nothing changes — then apply to the exact files listed.

1. Open the policy and choose to enforce it.
2. Microsoft 365: paste a SharePoint/OneDrive share URL, "🔍 Dry-run · no change". Each file shows aligned / wrong label / no label.
3. "✓ Confirm · apply to N files".
4. Google Drive: choose revoke external shares / anonymous links / everything except owners → dry-run → "Apply · revoke N shares".

Sweep the folder, choose what to remove, remove it.

1. Enter the folder and "🔍 Scan folder".
2. Pick a mode: list only · "anyone with the link" · all links · external access · "specific users" links · one specific user (by email).
3. "🚫 Revoke N permissions now" — or revoke row by row.

The set of rules a file travels with — lifetime, opens, device binding, watermark, burn. Policies choose envelopes; you rarely build one.

1. Seven standard envelopes, from light to strict: Public · Tracked · Painted link · Painted file · Hardware vault · Corporate (AIP) · View-once.
2. The composer’s rules pick one per situation — e.g. "external recipient + finance → View-once".

One by one with a role, or import from your Microsoft / Google directory.

1. "+ Issue new user": email, display name, a password of 10+ characters, and a role.
2. "Create user and issue password". The employee must set up 2FA at first sign-in.
3. Many at once: "Load from Microsoft 365" / "Load from Google Workspace", select people, choose a role.

Each employee has an allocation: which envelopes they may seal with, and their ceilings.

1. Organization members → the employee’s row → "Manage →".
2. Envelopes allowed for sealing — e.g. allow View-once only.
3. Ceilings: max lifetime in days, max opens, device binding (required / optional), sends per day, opens per day.
4. Write the reason, then "✓ Update allocation".

Write a composer rule with the condition "sender department" or "sender role".

1. "+ New rule" → condition "sender department" equals finance (or "sender role" equals contractor).
2. Add more conditions if needed (external destination, classification…).
3. Choose the envelope, "Simulate", "Save & enforce".

"Disable employee" cuts every tool and every data action instantly — and it is reversible.

1. Tools for employees → the employees table → "⛔ Disable employee".
2. Confirm: "All tools and all data access will be cut off instantly. This can be undone."
3. Changed your mind? "↻ Re-enable".

Tool tiles at the top of "Tools for employees".

1. Each tile toggles "On ✓" / "Off" for the whole organization.

Disable, revoke allocation, revoke devices, revoke what they sent — and only then, if you must, burn the user.

1. 1 · "⛔ Disable employee" — immediate and reversible.
2. 2 · Employee card → "✕ Revoke allocation (REVOKE)" — every sealing action is blocked.
3. 3 · Seal devices → "Revoke device" for each of their computers.
4. 4 · Vault → search what they sent → "🔥 Revoke" what should no longer open.
5. 5 · Only for a hostile exit: "Burn user" in the Courier hub — sessions, devices, password and 2FA are wiped. This cannot be undone.

Seal Studio. Choose how much control you need — you can still revoke any link-based share later.

1. Choose what should happen to the file: "Full control" (managed, per recipient — organizations), "Control the delivery" (a link you can revoke), or "Quick lock" (an offline file).
2. "📄 A file" or "🔑 Content or a secret" (paste text or a password). Drop it in.
3. Set the limits: expiry, max opens, view-only (no download), burn after first open, bind to first device, watermark.
4. Create it, then "📋 Copy link", QR card, WhatsApp or email.
5. Send any open code on a different channel than the link.

A request link: they upload, it arrives encrypted to you.

1. "Create a document request": what you need and for how long (24 hours to 30 days) → "Create request link".
2. Send the link. Send the OTP code separately.
3. When it arrives: "Get the file" decrypts it on your device.

Yes — the secure channel, with a delivery proof.

1. "📤 Send Document" or "📥 Request Document".
2. Pick the file (up to 10 MB), set lifetime and max views, "Encrypt & Send".
3. Share the link; send the access code separately.
4. Later: "Download evidence pack" — proof of delivery.

A quorum: k of n approvers must sign before the key is released.

1. Define the resource, how many signatures are required (k) out of how many approvers (n), and for how long.
2. Each approver signs from their own device.
3. Only after k signatures can the document be opened — once, for a short window.

"My deck" — shared with me, what I sent, my account and proofs.

1. My deck shows the status of everything you sent and received.
2. Inbox: All / Inbound / Outbound / By contact / Proofs, with "📂 Open" and "🧾 Proof" on each row.
3. Security, 2FA and passkey: My deck › Security.

Just open it. The file is decrypted locally in your browser.

1. Open the link. You will see "decrypting locally" and a signature check.
2. If it asks for a code — the sender sent it to you separately (SMS, a call).
3. Got a .html file? Save it, open it in a browser (on a phone use Chrome, not inside WhatsApp), enter the code.
4. "Has been revoked / burned" — the sender took it back. "Has expired" / "maximum opens" — its limits ran out. "Bound to another device" — open it on the device you first used.

Revoke it from the same screen. The next open fails.

1. On the result card: "🚫 Revoke this link".
2. It asks "Tap again — this is permanent". Tap again within a few seconds.
3. "✓ Revoked". Anyone opening it now sees "This share has been burned", and a page that is already open switches to it within seconds.

Everything you sealed is in the vault. Find it and revoke.

1. Vault → "🔎 Search a document by name…".
2. On the row: "🔥 Revoke" → "🔥 Sure? Click again".
3. Or in the Sharing control center: filter "Active", then "🔥 BURN" / "↩️ Revoke access" and confirm.
4. Open the share to see exactly who opened it, when, and what was refused.

Revoke that recipient only; everyone else keeps access.

1. Seal home → the file → the viewer’s row → revoke.
2. Confirm "Revoke access for …? Others keep their access." — you see "recipient revoked".

No — swap the code and keep the same link.

1. Vault → the file → "🔄 Rotate code".
2. Prove it is you: the old code, or "🔑 Forgot — 12 words", or the recovery key.
3. "🔄 Rotate code now" — the old code stops working, the link stays the same.
4. The link itself leaked? "🔗 New open link" — the old link stops working.

Revoke the device; then revoke anything sensitive it sent.

1. Seal devices → the laptop → "Revoke device" and confirm.
2. Then Vault → revoke files it sent that should not stay open.

Rotate it — the old key dies the same second.

1. API console → "Issue a key to paste". It replaces the current key immediately.
2. Update the key wherever it is configured.
3. Leaked an agent’s grant token instead? Revoke the agent (see "cut an AI agent off").

One seal, or all of them.

1. One: nodata revoke <seal_id>. You get a signed receipt and a verify URL.
2. Everything: nodata revoke --all --reason "…" and type "revoke all" to confirm.
3. nodata burn <file> overwrites and deletes a LOCAL file only.

nodata revoke sl_8f2c…
nodata revoke --all --reason "laptop stolen"
nodata policy            # who can reach what, right now

Most control actions are reversible. Destruction is not — that is what makes it trustworthy.

1. Disabled an employee → "↻ Re-enable" (then set their allocation again).
2. A wrong policy rule → edit or delete it in the composer and "Save & enforce" again.
3. Applied labels in Microsoft 365 → roll back within 60 minutes.
4. Revoked an agent → register it again and issue a new token.

Honest limits — know them before you share.

1. What someone already saw. A screenshot, a photo of the screen or a download already made stays with them. View-only PDFs carry a forensic serial so a leak can be traced back.
2. A "Quick lock" offline file, and a sovereign "Offline copy" once downloaded — by design they work without us, so we cannot recall them.
3. A revoke or burn itself. There is no "un-revoke": to share again, create a new link.
4. "Burn user" — the account is wiped.
5. Losing both the 12 words and the recovery file. We hold no copy of your keys.

Every open and every refusal is a signed receipt.

1. Per file: open the share — the event log lists every open and refusal.
2. For everything: Decisions — "who asked, what was decided, and what proves it": allowed, denied, granted, taken back.
3. Every row opens into its proof.

A CSV for the auditor; a signed ZIP for the lawyer.

1. Audit log → filter → "⬇ Download CSV".
2. Any single event → its incident page → evidence pack: a signed ZIP any lawyer can read.
3. Proofs → search by category (consents, communication, access, secrets).
4. Digest → a period summary for management: "Send to yourself / to the Board".

Verification is public, offline and open: signatures and a hash chain, checked in your own browser.

1. Drag an evidence pack (JSON) onto /verify-pack — nothing is uploaded.
2. Or paste a proof number (ND-2026-…) into /proof-factory and "Check".
3. From the terminal: nodata verify <proof_id>.

Read the forensic serial off the image and trace it.

1. Find the ND-… serial printed on the leaked page.
2. Paste it into the leak tracer (signed in to your organization).
3. "✓ Serial recovered" → issued to whom, which document.