Obsidian FIXED My OpenClaw Memory Loss!

Summary

In this video, Ron from BoxMining AI walks you through how he solved a major frustration with his OpenClaw AI agent — memory loss between sessions. Instead of relying solely on memory embeddings and vector indexing (which work like a black box), he found a more transparent and personal workaround: saving conversation logs as plain text or markdown files, storing them in Obsidian, and backing them up on GitHub. This combo acts as a second brain for your agent, letting it build a knowledge graph over time and connect the dots between past conversations. The more you use it, the better your agent gets at mimicking your voice and style — with noticeable improvements reportedly showing up by weeks four and five. Ron also shares two specific Obsidian plugins you should install: Smart Connections (for local semantic search and embeddings with no API key required) and QMMD as MD (for structured note formatting your agent can reliably write to). Beyond memory, he covers his two-part research workflow: a daily 7AM cron job that pulls crypto, market, and AI news via Brave Search API and drops a two-minute briefing into his Obsidian vault, and a deeper research mode that launches parallel sub-agents for more thorough analysis. He also explains why using Discord channels for your agent is far superior to WhatsApp or Telegram — focused topic channels keep the context window clean and improve output quality. Finally, Ron touches on using OpenClaw as a trading research assistant, emphasizing that you need your own real trading journal with actual trade data before your agent can help you identify patterns in wins and losses. There are no shortcuts here. He closes with a practical tip: always specify your time zone in your cron jobs, or your agent will default to UTC and run tasks at the wrong time.