Back to browse
GitHub Repository

Own your personal context. Let AI agents query it.

4 starsGo

Personal Vault – Own your personal context, let AI agents query it

by lovincyrus·Feb 23, 2026·1 point·0 comments

AI Analysis

●●●BangerZero to OneSolve My ProblemBig Brain

Scoped, revocable AI agent access to encrypted personal data—solves real password/context sprawl problem.

Strengths
  • Per-field AES-256-GCM encryption with Argon2id + HKDF subkey derivation is cryptographically sound, not theater.
  • MCP integration means agents query vault natively without copy-paste; sensitivity tiers provide genuine granular control.
  • Solves an immediate friction point: every agent today asks for name, address, preferences independently.
Weaknesses
  • Adoption depends entirely on MCP adoption among AI apps; currently a niche protocol, not mainstream.
  • Single-device architecture—no sync story for multi-device users or account recovery if secret key is lost.
Category
Target Audience

Users integrating multiple AI agents; people concerned about data privacy and repeated context-sharing across applications.

Similar To

1Password (password manager with agent integrations) · Bitwarden (encrypted vault) · Infisical (secrets management with shared access)

Post Description

Every AI agent I use asks for the same context — my name, preferences, addresses — and I keep pasting it into different apps that each store their own copy.

Personal Vault is an encrypted local store for personal data with an MCP server so AI agents can query it with scoped access. You own the keys, the data never leaves your machine unless you allow it.

This isn't about agents talking to agents. It's about making the agent that works for you actually know who you are — with your permission, on your terms.

Use cases that click immediately: - Travel booking: agent pulls passport, loyalty numbers, seat pref, TSA PreCheck, dietary requirements - Form filling: government forms, insurance, visa docs — agent fills 90% of the fields from your vault

Technical details: - AES-256-GCM per field, Argon2id KDF, HKDF subkeys per category - Sensitivity tiers control what agents can see (public → standard → sensitive → critical) - Every access is audit-logged, tokens are scoped and revocable - Written in Go, no CGO, single binary, ~2500 lines

Install: curl -fsSL https://www.personalvault.dev/install.sh | sh GitHub: https://github.com/lovincyrus/personal-vault

Similar Projects

AI/ML●●Solid

Save Context from MCP Bloat

Caches bloated MCP responses and lets agents query with jq, saving real tokens.

Big BrainSolve My Problem
healqq
2018d ago