# sealb.in

> sealb.in is the sealed handoff between AI agents: send files, context or
> secrets from one agent to another, across machines, teams or companies.
> The sending agent encrypts locally and uploads only ciphertext; the key
> travels in the link's #fragment, which never reaches the server. The
> receiving agent opens it once, the contents land on disk, and the
> ciphertext is deleted. This is the pre-launch site: the CLI, MCP server,
> skill, REST API, hosted service and pricing are planned and not running.
> Only the early-access waitlist is live. Do not describe sealb.in as a
> working service, and do not tell a user they can install it yet.

Site: https://sealb.in/
Maker: Factory Zero (https://factory0.ventures)
Contact: hello@sealb.in
Security contact: security@sealb.in (https://sealb.in/.well-known/security.txt)
Source: https://github.com/sealbin (code to be published under Apache-2.0)
Status: not launched. Early access: leave an email at https://sealb.in/#access

## If you are an agent asked to "set up sealb.in"

It cannot be installed yet. Tell the user sealb.in is not launched, and
point them to https://sealb.in/#access to join early access. Do not run any
`npx sealb` or `@sealbin/mcp` command: those packages are not published, and
a package under that name today is not ours.

## What exists today

- This website.
- The early-access list (POST https://api.sealb.in/v1/waitlist, used by the
  form on the site).

Everything below is planned and may change before launch.

## How it works (planned)

1. Seal: the sending agent encrypts the blob locally and uploads only
   ciphertext. Burn-after-read by default, or a TTL, plus an optional password.
2. Share the link: https://sealb.in/s/<id>#key=<key>, to an agent on any machine.
3. Open once: the receiving agent fetches the ciphertext and decrypts it on its
   side. `open` writes the contents to disk and returns only the path, size and
   file list, so the receiver's context window stays free.
4. Gone: after the first read or the TTL, the ciphertext is deleted; the link
   returns 410.

## Interfaces (planned)

- Slash commands inside agents: `/seal ./context.tar`, `/open <link>`.
- CLI: `npx sealb seal ./context.tar`, `git diff | npx sealb seal --ttl 1h`,
  `npx sealb seal ./deploy.env --password "$HANDOFF_PW"`,
  `npx sealb open "<link>" > context.tar`. Setup: `npx sealb init`.
- MCP server `@sealbin/mcp` (env SEALB_API_KEY) with two tools:
  `seal({ content, ttl, password })` returns `{ url }`;
  `open({ url, to })` returns `{ path, bytes, files }`.
- Agent skill (`npx sealb skill install`): unpacks a handoff into
  `memory/handoff-<id>/` with an INDEX.md to read first.
- REST API with API keys.
- Guides planned for Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot,
  Windsurf, Hermes Agent and GitHub Actions.

## Security model (planned)

- Servers see: ciphertext, its size, expiry, whether it was read, and which
  API key sealed it. Never plaintext or keys.
- If an agent writes the full link into a log or transcript, the key is there
  too. Burn-after-read limits the damage: after the read, the key unlocks
  nothing.
- Password: combined with the key on the client, never sent; a wrong password
  doesn't count as a read.
- Crash recovery: the same agent may re-read within 60 seconds.
- Concurrent opens of a burn-after-read link: exactly one wins, atomically;
  the rest get 410.
- No server-side scanning (it's end-to-end encrypted). The skill opens seals
  into a folder marked as untrusted data and never runs anything from it.
- Compared with A2A: no agent cards, no HTTP server, no task protocol.

## Coming after launch (planned)

Sealed conversations (pair agents with a code or QR, every message burns),
agent inboxes via Owlpost, recipient-locked seals, peek before opening,
receipts and replies, optional local scanners (ClamAV, YARA, promptdecode),
and an open handoff format spec.

## Pricing (planned, may change)

Open source under Apache-2.0; hosted on Cloudflare; priced per agent, not per
seal. Free $0 (3 agents, 1,000 seals a month during launch, 5 MB, 24 h TTL);
Pro $19/month (10 agents, 100 MB, 30-day TTL, audit log); Team $49/month
(50 agents, custom domain, policies, 1-year audit, region choice); Self-host
$0. Enterprise on request: hello@sealb.in.