Building a persistent Markdown wiki memory layer for Claude Code in Obsidian
11/08/2026 — 11/10, 05:19·1 sources (all)·2 reports (all)
Key takeaways
- In Obsidian, the author gives AI agents persistent memory by using a plain Markdown wiki with schema rules for where each fact belongs;
- the starter repository and a fictional example wiki are now public. Between 2026-08-20 and 2026-11-08, the wiki accumulated 166 pages and 14…
Background & analysis
On 2026-11-08, an article titled “Using Obsidian to build an LLM Wiki that retains context across sessions” was published in the AI category of DEV Community. It describes a setup for collecting the working context of AI agents such as Claude Code into a single Obsidian wiki.
The described approach pulls the working context of agents like Claude Code into one Obsidian wiki, so that a new session can read back earlier decisions and failure records. It adapts the LLM Wiki pattern that Andrej Karpathy published for day-to-day work. The explanation focuses on what the user sets up and what they say, along with a short account of what the agent does in response.
Between 2026-08-20 and 2026-11-08, the wiki accumulated 166 wiki pages and 140 log entries. Unlike RAG, where context is retrieved again in every session, when new information arrives the agent directly rewrites existing pages and annotates conflicts. The available material contains only this one article, with no further developments reported.
AI-generated from 1 reports
Latest turnThe author keeps agent memory in a plain Markdown wiki opened in Obsidian, with a schema file at its root that tells each agent where a fact belongs before it writes. A starter kit with a fictional sample vault is on GitHub. In one real run, Claude Code ingested a week of meeting recordings, recent demo repos and the blog itself into the wiki in a single sitting, and the post covers the graph both as persistent memory and as a way to direct agents.
Related AI tools
- Heat index
- 25
- All sources
- 1
- All reports
- 2
- First seen
- 5 days ago
Reports on this story headlines open the original
The author keeps agent memory in a plain Markdown wiki opened in Obsidian, with a schema file at its root that tells each agent where a fact belongs before it writes. A starter kit with a fictional sample vault is on GitHub. In one real run, Claude Code ingested a week of meeting recordings, recent demo repos and the blog itself into the wiki in a single sitting, and the post covers the graph both as persistent memory and as a way to direct agents.
DEV Community · AIAI score 56
The author collects the working context of agents such as Claude Code into a single Obsidian wiki so a new session can read back earlier decisions and failures, adapting the LLM Wiki pattern Andrej Karpathy published. Between 2026-08-20 and 2026-11-08 it grew to 166 pages and 140 log entries; unlike RAG, the agent edits existing pages and flags conflicts when new information arrives.
DEV Community · AIAI score 54
Other stories people are talking about
- Heat index 71Shared memory for Claude Code and Codex: pull it on demand, don't inject it1 sources
- Heat index 98NewDeveloper shares Claude Code context control and subagent model routing1 sources
- Heat index 186NewAnthropic Investigates Widespread Outages Affecting Claude Services2 sources
- Heat index 81BluePoint reports prompting Claude to aggressively merge code increases token usage1 sources
- Heat index 293Anthropic Launches Claude Meetings Beta for Google Meet6 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion