Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/coleam00/claude-memory-compiler/llms.txt

Use this file to discover all available pages before exploring further.

The knowledge base produced by Claude Memory Compiler is pure markdown. Every article uses Obsidian-style [[wikilinks]] and YAML frontmatter — the same formats Obsidian was built around. No export, no conversion, and no plugins are required. You can open knowledge/ as an Obsidian vault today and immediately see the graph of everything Claude has compiled from your conversations.

Why this works natively

All articles written by compile.py follow the conventions from AGENTS.md:
  • Wikilinks use the format [[path/to/article]] without the .md extension — exactly what Obsidian expects
  • YAML frontmatter contains title, tags, aliases, sources, created, and updated fields that Obsidian reads natively
  • File naming uses lowercase kebab-case (e.g., supabase-row-level-security.md) which is compatible with Obsidian’s link resolution
  • Articles link back to related articles via [[concepts/related-concept]] entries in the “Related Concepts” section, creating the backlink graph automatically

Setting up the vault

1

Open Obsidian

Launch Obsidian. If this is your first vault, you will be prompted to open or create one. If you already have Obsidian open, go to File → Open vault (or use the vault switcher icon in the bottom-left sidebar).
2

Point the vault at the knowledge/ directory

When prompted to select a vault location, navigate to your claude-memory-compiler clone and select the knowledge/ subdirectory. Obsidian will open that folder as the vault root.If you already have a vault and want to add the knowledge base as a second vault, use Open another vault from the vault switcher and point it at knowledge/.
3

Enable graph view

Open the graph view with Ctrl+G (or Cmd+G on macOS), or click the graph icon in the left sidebar. Obsidian will render every article as a node and every [[wikilink]] as an edge. Concept articles cluster by domain, and connection articles appear as bridges between clusters.

What you get

Once the vault is open, Obsidian provides:
  • Graph view — a visual map of every compiled article and the relationships between them. Clusters reveal how your knowledge is organized by domain.
  • Backlink panel — for any open article, Obsidian shows all other articles that link to it. This makes it easy to trace how a concept propagates across your knowledge base.
  • Full-text search — Obsidian’s built-in search (Ctrl+Shift+F) searches across every article in knowledge/, including the YAML frontmatter fields like tags and aliases.
  • Local graph — open any article and switch to local graph view to see only that article’s direct neighbors, useful for exploring the context around a specific concept.
Wikilinks are written without the .md extension by convention (e.g., [[concepts/supabase-auth]], not [[concepts/supabase-auth.md]]). Obsidian resolves these links correctly by default. Do not add .md extensions to wikilinks in your articles — query.py and lint.py parse links using the same extensionless format, so adding extensions would break the structural health checks.

The knowledge base is LLM-owned

Obsidian is read-mostly for this knowledge base. compile.py is the authoritative writer of articles in knowledge/concepts/, knowledge/connections/, and knowledge/qa/. You can read, browse, and search in Obsidian freely, but avoid hand-editing compiled articles — the next compile of the source daily log may overwrite your changes, and hand-edited content is not tracked by state.json. If you want to add permanent notes or project-specific context, add them to the daily/ log directly (in the AGENTS.md format) and let compile.py promote them into the knowledge base on the next run.
At approximately 2,000+ articles, the knowledge/index.md file becomes too large to fit in the context window used by query.py and session-start.py. At that scale, the index-guided retrieval approach hits its limit and you should consider adding a hybrid RAG layer (keyword + semantic search) as a pre-filter before the LLM. Until then, the structured index approach is simpler, cheaper, and more accurate for personal-scale knowledge bases.

Build docs developers (and LLMs) love