Tried AgriciDaniel/claude-obsidian: A Markdown-Native Second Brain for Claude Code

AgriciDaniel/claude-obsidian is an open-source workflow that turns Obsidian into a self-organizing AI second brain. Drop in a paper, webpage, transcript, or rough note, and Claude Code reads it, creates useful links, and files the result into a connected knowledge graph built from plain Markdown.

That “files it for me” layer is the compelling part. Instead of maintaining a rigid folder taxonomy—or locking valuable context inside a proprietary workspace—you keep the source files locally, searchable, versionable, and portable. The project follows the spirit of Karpathy’s LLM Wiki pattern while pushing it toward practical personal knowledge management: AI note-taking, automatic linking, and an open-source Notion alternative.

It is also gaining attention quickly: the repository added 631 stars today.

For teams already using an OpenAI- or Anthropic-compatible gateway, the integration can remain straightforward. For example, configure Claude Code to route requests through B-Lost’s Anthropic-compatible endpoint:

export ANTHROPIC_BASE_URL="https://b-lost.com/v1"
export ANTHROPIC_AUTH_TOKEN="$B_LOST_API_KEY"
export ANTHROPIC_MODEL="claude-fable-5"

claude

The exact environment-variable names may vary with your Claude Code setup, but the architecture is simple: Obsidian owns the Markdown vault, Claude Code performs ingestion and organization, and the gateway handles model routing and quota governance.

B-Lost’s native Anthropic /v1/messages support is particularly relevant here because repeated vault instructions and project context can benefit from full prompt caching, with cache hits discounted by 90%. That can make iterative note processing more predictable in both latency and cost. The relay also advertises 20% off official list pricing, while keeping the client-side workflow compatible with tools such as Cursor, Cline, Roo Code, Windsurf, Aider, and LibreChat.

For developers who want AI assistance without surrendering ownership of their knowledge base, this is a project worth watching.