OpenClaw + Obsidian gives you super powers
TL;DR
OpenClaw’s default memory misses the small stuff, so Alex adds Obsidian as a third memory layer — he frames it as the missing layer between built-in prompt memory and session history, giving agents a persistent vault they can search without bloating every prompt.
Daily logs, mistake tracking, and working context make the agent feel dramatically more reliable — Alex’s setup automatically writes high-level daily summaries, logs errors when he calls them out, and keeps dynamic project context so the agent can recover what happened before a compaction.
The big win is memory-on-demand, not dumping everything into context — instead of injecting the whole vault into every prompt, the agent checks Obsidian at session start and pulls the exact note it needs, like a project from 3 days or even 3 months ago.
Compaction stops feeling like a brain wipe once the agent can re-read its own recent notes — Alex says OpenClaw used to forget things ‘from 5 seconds ago,’ but with the vault, it can look up what happened right before compaction and continue without him noticing the reset.
A shared Obsidian workspace turns multiple agents into a real team — Alex uses Hermes and OpenClaw side by side, and because both can read the same shared folder, one agent can pick up a YouTube video project the other agent started immediately.
Setup is mostly a copy-paste prompt, but you need to verify the rules actually stick — after installing Obsidian and pointing the prompt to the vault path, Alex warns that weaker models may fail to write the new behavior into agents.md, so you should check that memories are actually being saved.
The Breakdown
The core pitch: OpenClaw’s memory ‘sucks,’ but Obsidian fixes it
Alex opens bluntly: OpenClaw forgets too much, and he says he’s built a system with Obsidian that makes its memory ‘basically flawless.’ He immediately previews the payoff: this isn’t just for OpenClaw, but for Hermes or really any AI agent you want to give ‘super powers.’
Inside the vault: daily logs, mistakes, and shared memory
He tours his Obsidian setup like a workspace reveal. Each agent gets its own area, with automatic daily logs for everything discussed, a mistakes file that records every time the agent messes up, and a working-context file for whatever the agent needs right now. The especially useful bit is the shared workspace, where all agents can see the same project knowledge.
Why Obsidian works so well for agents
Alex takes a beat to explain Obsidian for anyone new: it’s a free markdown interface, basically a clean home for text files. His bigger point is that markdown has become ‘the language of AI,’ and since OpenClaw already runs on markdown files, Obsidian is a natural bridge between human organization and agent-readable memory.
The four-layer memory model
He breaks OpenClaw’s memory into layers: built-in memory for the essentials, agents.md for rules, soul.md for personality, and session search for historical sessions and cron jobs. The new ingredient is layer three — the Obsidian vault — which holds important long-term context without forcing all of it into every single prompt.
The compaction problem, and why this setup changes everything
Alex says the real pain point is compaction, when OpenClaw can suddenly feel ‘stupid’ and forget what happened seconds ago. With the vault in place, the agent checks what was happening right before compaction and reloads that context, which he says makes compaction almost invisible. His example is practical: now he can say, ‘Hey, let’s work on that project from 3 days ago,’ and the agent can actually find it.
Shared folders make multi-agent workflows click
This is where the system expands from memory to coordination. Alex describes scripting a YouTube video with Hermes, then switching to OpenClaw and having it instantly continue because the project lived in the shared directory. His takeaway is clear: if you’re only using one agent, you should probably be using at least two once they can share context this way.
Setup is simple, but don’t skip the rule-check
The implementation is refreshingly low drama: install Obsidian, paste in his prompt, and let the agent build the system. The one caution is that some weaker models won’t properly write the workflow into their own agents.md rules file, so Alex tells you to verify that notes are actually being saved and, if needed, explicitly tell the agent to ‘burn’ the system into its rules.
Beyond memory: your own personal AI wiki
He closes by showing the next layer of ambition: once everything is stored in Obsidian, you can build interfaces on top of it. Alex says he asked OpenClaw to create a visual ‘Alex Finn wiki’ of people, concepts, and sources from his memory, turning the vault into something browsable and much more useful than just a pile of notes.