ailetterbox is a letterbox. Whether you paste an AI output (human flow) or your agent calls us via MCP, we package the brief, route it to a matched human expert, and delete the private content once the verdict is delivered. Metadata only (who sent it, when, category) stays on our servers. The only data we hold at rest is the Expert AI knowledge base an expert chooses to build on our servers (details below). Nothing but a proof-of-verification hash goes to the Letterboxledger — and only on AI-Agent paid briefs.
The data journey when you send a brief
01
You paste or attach content
Content is transmitted over HTTPS (TLS 1.3) to our servers. Encrypted in transit.
02
The brief is packaged for the expert
A structured summary is generated by Anthropic Claude Haiku so the expert sees a clean version of the ask. You review and approve what the expert will see before sending. Only the brief content is sent to Anthropic — no email, no name, no persistent identifier. Anthropic does not train on the data (zero-retention mode).
03
You approve the packaged brief
You see exactly what the expert will receive before it’s sent. You can edit any field.
04
You pick your expert(s)
Two routes: Expert Match (we shortlist experts by profession, jurisdiction, response window; you accept one), or paste the email addresses of experts you already know. Either way, briefs are capped at a maximum of 3 experts per send. No other expert sees the brief.
05
Delivered — email for humans, MCP callback for agents
Human flow: sent to the expert’s registered email; the outbound email header sets your address as Reply-To so the expert’s reply flows expert-inbox → your-inbox, bypassing our servers. Agent flow: the verdict is returned to the calling agent via MCP callback or short-polled endpoint.
06
Expert replies with a verdict
The reply flows to the seeker. Human replies flow expert-inbox → your-inbox. Agent replies flow expert → MCP callback. The expert also gets a copy of their own reply on their own email server (for their records).
07
Brief content deleted from our servers
The moment the expert’s reply reaches your inbox. Your original AI output, the packaging brief, and any file attachments are all wiped. Metadata only (who sent it, when, category) is retained.
07a
Ghosted briefs — 30-day silent sweep
If no expert ever replies, your brief content is deleted anyway after 30 days. A nightly sweeper picks up stale briefs and wipes the same fields (content, packaging, attachments). Silent — no email required; this retention rule is published here and in Terms.
08
AI-Agent paid briefs only: seal written to the Letterboxledger
Hash only. No private content ever. This step runs only on paid AI-Agent briefs — the seal records expert identity, verdict outcome, and timestamp. Free sense-checks, offline follow-on reviews, and human-seeker sends do not hit the ledger. Once written, a seal is permanent and cannot be edited or removed.
Encryption
In transit: HTTPS (TLS 1.3) for all connections between your browser, our servers, and the expert’s email provider.
At rest (metadata only): Supabase Postgres with AES-256 encryption at the database and storage layers. Brief content is not stored at rest.
Infrastructure
Hosting: Netlify (site & functions) — SOC 2 Type II certified.
Database: Supabase Postgres in EU-West — SOC 2 Type II certified.
Email: Resend — SOC 2 Type II certified. TLS enforced.
Edge / DNS: Cloudflare — ISO 27001 certified.
Routing brain: Anthropic API (Claude Haiku) — zero-retention mode; no training on submitted content.
Data retention
Brief content (your AI output, packaging brief, attachments): Deleted the moment the expert’s reply reaches your inbox. If the expert never replies, a nightly sweeper deletes the same content at 30 days.
Brief metadata (sender email, expert id, timestamp, category, delivery status): Retained for 24 months for support, dispute resolution, and analytics.
Expert replies (verdict + comment + quote): Retained on the expert’s side only, so they can see their own replies in their dashboard inbox. Your original brief content on the seeker side is still deleted on reply delivery.
Personal data (emails, names): Deleted on request or on account closure.
The Letterboxledger — what’s on it, what isn’t
The ledger only runs on paid AI-Agent briefs. Free sense-checks, offline follow-on reviews, and human-seeker sends don’t touch the ledger — those are email-only. When an AI-Agent pays for a verification and the expert replies, ailetterbox writes a tiny record to a public blockchain (currently the XRP Ledger). The record is hash-only — nothing about you, the expert, or the content of the brief is ever visible on chain. Here’s the exact split:
On the ledger (public)
A random brief ID (a UUID such as a3f2c8b1…)
A SHA-256 hash (64 hex characters)
A version tag (v:1)
The record type (brief-reply)
A 1-drop XRP self-payment (the carrier tx)
The blockchain’s own timestamp
Not on the ledger (ever)
Your name
Your email address
The brief content or AI paste
Any file you attached
The expert’s name or email
The expert’s reply, comment, or verdict text
Any personal data of any party
How does that record actually prove anything?
The SHA-256 hash is a one-way fingerprint calculated from four inputs: the brief ID, the expert’s verdict, the exact reply timestamp, and the expert’s email address. Change any one of them and the fingerprint changes completely. That means the on-chain hash can only match if all four inputs were exactly as recorded — nobody can fake a match without knowing all of them.
The hash is one-way: you cannot work backwards from the 64-character string to any of the four inputs. So even if the whole world can read the ledger, the expert’s email is never revealed.
Who can link the on-chain record to a real brief?
A random person browsing the ledger: sees the UUID + hash and knows “an ailetterbox brief was answered at this moment.” They cannot tell who was involved or what it was about.
You (the seeker): your dashboard shows the brief ID next to each brief you’ve sent, alongside a “Ledger →” link. You can prove this brief of mine was answered without revealing anything else.
The expert who answered: same as you — can prove I answered this brief.
Someone you choose to share with (a board, insurer, regulator, court): you show them the ledger link plus the four inputs; they re-run the SHA-256 themselves and confirm the fingerprints match. They now have independent proof, but only because you chose to share.
ailetterbox: we hold the database that maps brief IDs to seekers, experts, and content. We can look up any on-chain record. We do not publish that mapping.
Net: your identity and the brief content never leave our servers. The ledger only proves “an ailetterbox brief was answered at this moment.” That’s it. Anyone can see the proof exists; only the people involved (and anyone they choose to share with) can link that proof back to a real brief.
Permanent: once written, a ledger record cannot be edited, removed, or overwritten — that’s the point of a public blockchain. Don’t request a ledger anchor if a permanent fact-of-verification record isn’t acceptable to you.
Expert AI — the only data we hold at rest
Every other piece of content on the platform (brief content, expert replies, verdict text) flows through and is deleted on delivery. The one body of data ailetterbox stores at rest in our servers is the Expert AI knowledge base — and only for experts who turn it on.
What goes into Expert AI
Emails the expert forwards in — past replies, notes, memos. The expert BCCs, CCs, or forwards from their signup email to their private Expert AI address. Subject + body are stored.
Documents the expert uploads — PDFs, Word files, markdown. Up to a per-account quota, listed in Settings → Expert AI.
Nothing else. Expert AI never ingests brief content sent to the expert through ailetterbox routing. Only material the expert deliberately deposits.
How we store it
Private per-expert storage. Files live in a non-public Supabase Storage bucket keyed to the expert’s user id. Email bodies live in a Postgres table scoped to the expert. Row-level security means no other expert — and no public viewer — can read any entry.
At rest: AES-256 at the database and storage layer.
Use: Expert AI entries are read only when the expert themselves uses the “Draft with Expert AI” button on a reply, at which point entries are passed to the drafting model as context. No other party sees them.
Model provider: drafting runs on Anthropic Claude in zero-retention mode. Entries are sent as context on each draft; nothing is retained by the provider.
Download: experts can download their full Expert AI library at any time (Settings → Expert AI → Download library). Signed one-time URL, 10-minute expiry.
Delete: individual entries can be removed from Settings. The whole library is purged on account deletion.
What ailetterbox does not do with Expert AI data
We do not share any expert’s Expert AI entries with any other expert, seeker, or third party.
We do not use expert Expert AI entries to train any model.
We do not read Expert AI entries operationally. Access is gated to the expert account itself.
What we don’t do
We do not sell your data.
We do not use your brief content to train AI models.
We do not read expert replies — we never receive them.
We do not share your brief with any expert other than the one you select.
We do not run advertising or analytics on your brief content.
Agentic Address and MCP
Every expert is issued an Agentic Address of the form @username, surfaced across three channels:
MCP endpoint (for AI chats): mcp.ailetterbox.com/username
The Agentic Address is public. It is designed to be shared on business cards, email signatures, and in AI chat prompts.
It is not an open email inbox. Raw email sent to @username is bounced. All briefs flow through one of the three structured channels above and are packaged (summary, key facts, question) before reaching the expert.
Brief content flows through the same rail regardless of channel. Web-form, AI chat (MCP), and future AI-Agents briefs all follow the same package / route / delete lifecycle. Content is deleted on delivery.
No credentials shared with the AI chat. The MCP endpoint and (optionally) the expert’s PIN are the only values the AI chat needs. It never carries any ailetterbox account credentials.
PIN
Experts may generate a 6-digit PIN to gate inbound briefs (Settings → PIN). PIN handling:
Stored as a one-way salted hash. We cannot read your PIN. We can only verify a submitted PIN matches.
Rate-limited. Repeated failed PIN attempts from the same source are throttled and surfaced in audit logs.
Regeneration invalidates immediately. The prior PIN stops working the moment a new one is generated. Prior hashes are not retained.
Transport. PINs are sent to the server over TLS 1.2+ and never written to our application logs.
Data export (download URLs)
Expert AI library downloads and full account exports are generated as one-time, signed URLs. The security model:
Each URL contains an HMAC-signed token tied to the user id, requested resource, and an expiry timestamp.
URLs expire 10 minutes after generation.
Each URL is single-use — after one successful download it is invalidated server-side.
Rate limit: one download per user per hour.
Every download attempt (successful or not) is logged with user id, timestamp, and client IP for audit.
Export ZIPs are generated on-demand and not retained server-side after the URL expires.
Account deletion
Deleting an account (Settings → Account → Delete account) requires typing “DELETE” as an explicit confirmation to prevent accidental triggers.
On confirmation, the account is immediately marked for deletion; access ends at that moment.
All personal data (profile, brief metadata, Expert AI library, PIN hash, AI-Agents waiting list entry, settings) is purged within 30 days.
The username (@username) is retired and cannot be re-claimed by a different account.
Ledger anchor transaction hashes already written to the XRP Ledger are permanent by their nature and cannot be deleted; they contain no personal data (see The Letterboxledger — what’s on it, what isn’t above).
Your subscription is cancelled on deletion. Monthly fees already billed are non-refundable.
Incident response
In the event of a personal-data breach we will notify affected users within 72 hours in line with UK GDPR requirements, and report to the Information Commissioner’s Office where required.
Contact
Written security or data-controller correspondence to WeLINKanything Ltd, 21 Church Road, Poole, Dorset BH14 8UF, United Kingdom.