VaultAgent

VaultAgent is a macOS application described by its author as a privacy-first personal vault for local document indexing and AI query processing. The product claims to enable users to ask questions about private documents without exposing sensitive information to cloud-based LLMs, using a local redaction gate before any data reaches an external model.

The author states that VaultAgent indexes PDFs, Office files, text/Markdown, email archives, WhatsApp exports, and image scans locally. It uses tools like Presidio for structured PII detection, Gemma 4 E2B for contextual PII, and Ollama for local processing. A key feature is the “redaction gate” that replaces sensitive data with typed placeholders such as [CREDIT_CARD], and a dashboard showing what was redacted.

The description indicates a single-founder team (Gurman Singh), built during an OpenAI hackathon, and includes self-reported performance metrics like 525/525 hard redaction stress cases clean, 0 leaks in real-audit replay, and GPT-5.6 Terra answering 9/10 fixed vault questions.

This analysis is based entirely on self-reported information from the project description provided by the caller — no external verification, no revenue data, no customer names, and no third-party sources. All claims are treated as stated by the author, not proven.

The description states that VaultAgent is a privacy-first personal vault for macOS, designed to index documents locally and allow users to query them using AI without exposing sensitive data to cloud models.

It indexes:

• PDFs
• Office documents
• Text/Markdown
• Email archives
• WhatsApp exports
• Image scans

The system processes queries through a local redaction gate before sending any context to an external LLM. It uses:

• Apple Vision OCR
• EmbeddingGemma
• SQLite with FTS5 and sqlite-vec
• FastAPI
• Ollama
• Presidio
• FastMCP

It also supports:

• Tunnel-friendly read-only MCP for ChatGPT
• Localhost-only MCP for trusted Codex file imports

The author describes the tool as being built using Codex as an engineering collaborator, and includes performance metrics such as:

• 525/525 hard redaction stress cases clean
• 0 leaks in a real-audit replay of 1,786 fields against 8 canaries
• GPT-5.6 Terra answering 9/10 fixed vault questions

Inference The product is described as a local-first AI assistant for personal documents with built-in privacy controls.

The author positions VaultAgent as a solution to the problem of AI access to private data, especially when using cloud-based LLMs. The tagline says:

“Ever thought of using behemoth OpenAI LLMs without worrying about private info being shared accidentally? Now you can do worry-free uploads, Power and Privacy both go hand in hand now with VaultAgent.”

The description frames it as:

• A privacy-first personal vault
• A tool that allows users to ask questions about their own documents
• A way to avoid uploading raw records or identifiers to cloud models

It also emphasizes:

• Local processing
• Redaction of PII before any data leaves the device
• Visibility into what was redacted (dashboard shows exact local text beside redacted payload)
• Speed vs. privacy trade-off, with Fast and Full modes exposed

Inference The positioning evolves from a technical solution to a privacy-conscious UX for personal document AI interaction.

The description does not explicitly name target customers or define an Ideal Customer Profile (ICP). However, it implies:

• Personal users who store sensitive documents (medical, financial, etc.)
• Users concerned about data privacy when interacting with AI
• macOS users who want to query their own files locally

It is described as a personal vault for macOS — not a business or enterprise product.

Inference The target customer appears to be individuals with private documents they wish to query using AI, particularly those concerned with privacy and data control.

No evidence of pricing, monetization strategy, or business model is provided in the description. The author does not mention:

• Subscription plans
• Freemium tiers
• Licensing models
• Revenue streams
• Paid features

The project is described as a hackathon submission and built by one person.

Inference There is no evidence of a business model or pricing strategy, and no indication that VaultAgent has moved beyond prototype or personal use.

The author describes:

• A local indexing stack using Apple Vision OCR, EmbeddingGemma, SQLite with FTS5 and sqlite-vec
• FastAPI, Ollama, Presidio, FastMCP
• Codex as an engineering collaborator for building evaluation harnesses and testing
• Redaction gate that includes:
  • Microsoft Presidio for structured PII
  • Gemma 4 E2B for contextual/prose PII
• Dashboard showing redacted vs. local text
• Tunnel-friendly MCP for ChatGPT and localhost-only MCP for Codex
• Deletion-aware indexing, per-file removal/reset controls, authenticated MCP boundaries, opaque record IDs, audit trail

Performance metrics include:

• 525/525 hard redaction stress cases clean
• 0 leaks in real-audit replay
• GPT-5.6 Terra answering 9/10 fixed vault questions
• 44/44 regression tests pass

Inference The technical stack is described as robust and includes local processing, privacy controls, and auditability.

The description does not provide any evidence of:

• Revenue
• Customers or user base
• Product adoption
• Market traction
• Product maturity beyond prototype stage

It is described as a hackathon submission by one person (Gurman Singh), and the author states that it was built for the OpenAI 2026 hackathon.

Inference There is no evidence of traction or product maturity, and no indication that VaultAgent has moved beyond a proof-of-concept stage.

The description does not mention any competitors. It does not reference:

• Similar privacy-focused AI tools
• Document vaults with local processing
• LLM privacy gateways or redaction tools

It is framed as a unique solution to the problem of private document querying via AI, but no competitive landscape is described.

Inference No evidence of competitive context is provided. The product may be positioned in a niche space, but there is no indication of existing players or market dynamics.

• No commercial traction or revenue: The project is described as a hackathon submission with no evidence of users or monetization.
• Single-founder team: No team structure or scaling plan is evident.
• Unverified claims: All performance and privacy metrics are self-reported without independent verification.
• Limited scope: Designed only for macOS, personal use — not enterprise or broader market.
• No pricing or business model: No indication of how the product would be monetized.
• Unproven UX maturity: While the author mentions “privacy UX has to be visible,” there is no evidence of user testing or feedback.

Inference The main risk is that VaultAgent remains a prototype with no commercial viability or scalability beyond personal use.

1. What is your plan for scaling beyond a single-person hackathon project?
2. Have you validated the privacy UX with real users, or is it based on assumptions?
3. How do you intend to monetize this product if it remains a personal tool?
4. Are there any plans to expand beyond macOS or support enterprise use cases?
5. What are your long-term goals for VaultAgent — is it intended as a standalone product or part of a larger platform?
6. Do you have any feedback from users or early adopters about usability or privacy concerns?