Agent Knowledge Vault

An Agent Knowledge Vault is a shared, model-agnostic knowledge base built on plain Markdown (specifically yanp-for-agentic-workflows) that allows humans and multiple AI agents (Claude, Gemini, etc.) to collaborate on a single knowledge graph.

The Problem of Siloed Memory

Most agent memory systems are fragmented and specific to a single provider or tool. This vault acts as a layer above them, providing a unified space for:

• User preferences and workflow standards.
• Project state and design rationale.
• Cross-session continuity.
• Operational conventions.

Core Pillars

1. Interoperability: Built on standard Markdown and Git to ensure any tool can read/write.
2. Provenance: Clear metadata (author, hostname, date) to track who wrote what and where.
3. Auditability: Using Git for version control and rollbacks.
4. Graph-Based: Using Wikilinks to create a dense web of information rather than isolated files.

Operational Standards

The vault relies on strict adherence to agent-note-conventions to prevent drift and ensure that machine-parsable metadata remains consistent over time.

References

• Source: 00_Raw/agent-knowledge-vault.md
• core-patterns-moc
• wiki-as-codebase
• agent-note-conventions
• index
• llm-wiki-pattern

Persistent Memory Patterns

• adk-long-term-memory - Cross-session recall and knowledge persistence.