Perseus Vault Codex

The description states that Perseus Vault Codex is an MCP server that provides persistent, encrypted, local-first memory for OpenAI Codex agents. It integrates with Codex through a JSON-RPC 2.0 interface over stdio using the Model Context Protocol (MCP). It wraps a Rust-based single-binary database called Perseus Vault, which uses SQLite + FTS5 and AES-256-GCM encryption at rest.

It provides five memory verbs:

• perseus_remember: stores facts or decisions.
• perseus_recall: retrieves relevant context using keyword + hybrid recall.
• perseus_forget: soft-deletes stale memories.
• perseus_reflect: synthesizes insights from recalled memories using an LLM (fallback to context-only).
• perseus_status: reports memory count, encryption status, and location of local DB.

The system is designed to be zero-config: it auto-creates an encrypted vault at ~/.perseus-vault/codex/memory.db on first run and generates its own AES key. Integration requires one pip install and one setup command (perseus-vault-codex-setup). It does not require API keys, cloud dependencies, or telemetry.

Evidence The description states this is the product. No external validation or usage data provided.

The author positions Perseus Vault Codex as a local-first, encrypted memory solution for developer agents, specifically targeting users of OpenAI Codex who want to avoid re-learning project context across sessions.

It claims to be:

• A single-binary tool with no cloud dependency.
• Encrypted at rest using AES-256-GCM.
• Zero-infra, with no API keys or telemetry.
• Designed for developers who value security and privacy of unreleased code and architecture.
• An answer to the missing "memory primitive" in coding agents.

It contrasts itself with existing tools like mem0, cognee, Letta, and Chroma, stating that none offer a single-binary, zero-infra, encrypted solution.

Evidence The description is self-reported. No third-party validation or market positioning data provided.

The description states that Perseus Vault Codex targets developers using OpenAI Codex agents, particularly those who value:

• Security and privacy of unreleased code, architecture, and secrets.
• Local-first memory without cloud dependencies.
• Persistent context across sessions to avoid re-learning.

It is positioned for developers working in environments where data should not leave the machine unencrypted.

Evidence The description states this is the target audience. No customer or user data provided.

The project is described as a self-contained tool, built with Python and Rust, and integrated via pip install and setup command. It does not appear to have any pricing model or monetization strategy described.

There is no mention of:

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

Evidence Not evidenced.

The system is built with:

• Python (zero runtime dependencies)
• Rust-based single-binary Perseus Vault
• SQLite + FTS5 for local storage and search
• AES-256-GCM encryption at rest
• Model Context Protocol (MCP) over stdio
• OpenAI Codex and GPT-5.6 used during development

It includes:

• A 31-test suite
• End-to-end verification of integration with the real Perseus Vault binary
• Benchmarks: recall at p50 8 ms / 5-of-5 recall@10 on a 10k-memory corpus, and 72.5% token reduction over 30 sessions vs. re-priming each session

Evidence The description states these are the technical components and benchmarks. No independent verification or performance data beyond author claims.

The project is described as a hackathon submission, built during Build Week, and submitted to the OpenAI 2026 hackathon on Devpost.

It includes:

• A real integration with Codex
• A test suite of 31 tests
• End-to-end verification
• Measured benchmarks

However, there is no evidence of:

• Real-world usage or adoption
• Customer base
• Revenue
• Product-market fit
• Market traction beyond the author’s own testing

Evidence Not evidenced.

The description states that existing tools like:

• mem0: cloud-dependent
• cognee: Python-only, no encryption at rest
• Letta: does not encrypt local storage
• Chroma: vector DB, not structured agent memory

None of these are described as having a single-binary, zero-infra, encrypted solution like Perseus Vault Codex.

Evidence The description states this is the competitive landscape. No independent data or market positioning provided.

• No real-world usage or adoption: The product is described only as a hackathon submission with no evidence of traction.
• Self-reported benchmarks and claims: No third-party validation of performance or token savings.
• Single-person team: Only one member (Thomas Connally) is listed, raising questions about scalability and long-term maintenance.
• No pricing or monetization model: The product appears to be a tool for developers, but no business model is described.
• Limited scope: The project focuses on Codex integration and does not appear to have broader applicability or roadmap beyond this.

Evidence These are inferences based on the lack of evidence for traction, pricing, or team size beyond what is stated.

1. What is the current usage or adoption rate of Perseus Vault Codex?
2. How does the product plan to scale beyond a single developer tool?
3. Is there any intention to monetize or commercialize this tool?
4. Are there plans to support other agents beyond OpenAI Codex?
5. What are the long-term maintenance and security plans for the Rust-based Perseus Vault binary?
6. How is the encryption at rest validated in production?
7. Has the product been tested with real-world use cases or teams?