memvault

LLMs lose context between sessions, forcing you to repeat preferences, architectural decisions, and project paths every time you open a new chat. This skill solves the "goldfish memory" problem by integrating MemVault, a local-first memory engine that uses the Model Context Protocol (MCP) to persist knowledge across conversations without relying on cloud storage.

What it does

• Persistent storage saves specific facts, preferences, and decisions to a local database that survives session restarts.
• Context injection retrieves relevant historical data via keyword or vector search and feeds it directly into the current prompt.
• Fact versioning updates existing memories by superseding old facts with new ones while maintaining a verifiable history.
• Verified recall uses a hash-chained ledger to explain exactly why a specific fact was retrieved for a given query.
• Privacy control runs entirely on your local machine with no network access unless you explicitly configure an external embedding provider.

How it works

1. Installation downloads the MemVault binary and generates the necessary JSON configuration for your MCP client.
2. Namespace creation segments memories by project or repository name to ensure relevant retrieval without cross-talk.
3. Write loop captures stand-alone facts during the conversation, marking critical architectural rules as pinned.
4. Search loop queries the local vault at the start of new tasks to pull in previously learned context.

Frameworks & tools

This skill uses the Model Context Protocol (MCP) and interfaces with the MemVault Rust-based binary. It integrates with any MCP-compliant client such as Claude Code or Cursor and supports OpenAI-compatible embedding endpoints like Ollama for semantic search.

Why this beats prompting it yourself

Writing to a text file is messy and lacks ranking. This skill provides a structured API for the agent to manage its own memory, ensuring the most relevant facts are prioritized within the token budget while maintaining a permanent, auditable record of why information was recalled.

Use cases

• Store project conventions like specific linting rules or deployment paths that are not in the README.
• Remember user preferences regarding coding style, documentation depth, or preferred libraries across different sessions.
• Track architectural decisions and the reasoning behind them to prevent re-litigating the same technical debt later.
• Maintain stateful workflows by recording the progress of long-running tasks that span multiple days or chat threads.

Known limitations

Requires manual binary installation and client restart for initial setup. Semantic search requires an external embedding provider like Ollama; otherwise, retrieval defaults to keyword-only BM25.