George Orwell — 1984
We did. Your L2 key is derived in your browser from your Touch ID. Our servers have never seen it. They could not decrypt your private fields even if they wanted to. Or anybody else.
Every password manager was built before AI agents existed. Now they need to catch up.
All others give your AI agent access to everything in your vault, or nothing at all. There's no middle ground. Your AI needs your GitHub token — it shouldn't also see your passport number.
"AI-safe" vaults still decrypt everything server-side. They rely on access policies that can be overridden, misconfigured, or bypassed. If the server can read it, it's not truly private.
Your AI can't log in to a service, get past two-factor authentication, or rotate API keys without credential access. vault1984 lets it do all three — without exposing your credit card or passport to the same pipeline.
"Your assistant can book your flights.
Not read your diary."
Your passwords are stored on the vault server — yours to self-host, or ours to run. Every field is encrypted. But some fields get a second lock. That second key is derived from your fingerprint and only exists in your browser. The server holds the safe. Only you hold that key.
Encrypted at rest, decryptable by the vault server. Your AI agent reads these via MCP.
Encrypted client-side with WebAuthn PRF. The server never sees the plaintext. Ever.
Not another password manager with an AI checkbox. The architecture is the feature.
Each field in an entry has its own encryption tier. Your AI reads the username, not the CVV. Same entry, different access.
L2 encryption uses WebAuthn PRF — a cryptographic key derived from your biometric hardware. Math, not policy. The server literally cannot decrypt it.
Store TOTP secrets as L1 fields. Your AI agent generates time-based codes on demand via MCP — no more switching to your authenticator app.
Create separate MCP tokens per agent or integration. Each token sees only its designated entries. Compromise one, the rest stay clean.
No Docker. No Postgres. No Redis. One Go binary, one SQLite file. Runs on a Raspberry Pi. Runs on a VPS. Runs on your laptop.
Import from any password manager and the built-in LLM automatically classifies which fields should be L1 (AI-visible) vs L2 (private).
Create scoped MCP tokens per agent. One compromised agent exposes one agent's scope — not your entire vault.
{
"mcpServers": {
"vault-dev": {
"url": "http://localhost:1984/mcp",
"headers": {
"Authorization": "Bearer mcp_dev_a3f8..."
}
},
"vault-social": {
"url": "http://localhost:1984/mcp",
"headers": {
"Authorization": "Bearer mcp_social_7b2e..."
}
}
}
}We host vault1984 on TIER III infrastructure across four regions. $12/year. Pick your region at signup.
Your L2 keys are derived in your browser. We mathematically cannot read your private fields.
One command. No dependencies.
{
"mcpServers": {
"vault1984": {
"url": "http://localhost:1984/mcp",
"headers": {
"Authorization": "Bearer mcp_your_token_here"
}
}
}
}