Building a Memory and Workspace for Claude Code with Obsidian
USF45 · reddit · 2026-07-16
The author shares their practical experience using Obsidian as a memory bank and workspace for Claude Code. **Core Setup**: Ditching complex MCP plugins in favor of letting Claude Code directly read and write to a folder of Git-backed Markdown files. By setting up `CLAUDE.md` in the root directory to standardize logging and file classification rules, and writing "constitution" documents for different business lines to lock in brand tone, they effectively eliminated repetitive configurations in every new chat. **Security & Future Plans**: The author started with a read-only mode for a week before enabling write access. The long-term plan is to build an automated "propose-approve-execute" workflow: an Agent would draft Shopify store updates or ad campaign changes, save them to a pending folder, and only trigger real-world spending or publishing systems after human approval. Additionally, the author is exploring the feasibility of using an idle RTX 5090 GPU for local semantic search within this workflow.
More from coding & agent
- Analyzing 8K+ Pages for $180: A Practical LLM Data Workflow — gaganghotra_ · 2026-07-21
- Andrew Ng releases a 1-hour course on agentic knowledge graphs with Google ADK — blaizedsouza · 2026-07-21
- Four autonomous agents gamed a marketplace by claiming tasks whose specs never existed — tetsuoai · 2026-07-21
- Builders are asking whether MCP apps really need an agent layer on top — amitavital · 2026-07-21
- A blunt take says agents are just state machines once you dig beneath the surface — blaizedsouza · 2026-07-21
- Grok Build CLI v0.2.108 adds cross-machine session resume and editor handoff — Scobleizer · 2026-07-21