OpenClaw Keeps Forgetting Things — Here's the Permanent Fix Using Obsidian

One of the first things people discover after setting up OpenClaw — usually after a few weeks of enthusiastic use — is that the "persistent memory" isn't as permanent as advertised.

Conversations get compacted. Details drop out. An agent that knew everything about your projects last week can feel strangely amnesiac today.

This is a solvable problem, and the solution is connecting OpenClaw to Obsidian — a free, locally-stored note-taking app that gives your agent a genuine, searchable, structured long-term memory that it can read, write, and reason over.

Here's exactly how to do it.

Why OpenClaw's Default Memory Falls Short

To understand the fix, it helps to understand the problem.

OpenClaw maintains memory in a memory.md file that gets added to with every interaction. Over time, this file balloons. To manage token costs and context window limits, OpenClaw periodically runs compaction — summarising older conversations into shorter entries.

The problem: compaction is lossy. Details that felt important get summarised away. Specific project notes, personal preferences you mentioned in passing, nuanced context — it all gradually degrades. After a few months of use, your agent has a vague impression of your life rather than a clear understanding of it.

The Obsidian integration replaces this with something better: a structured vault of markdown files that your agent can read and write like a knowledge base. Nothing gets automatically compressed or discarded.

What You'll Need

Obsidian — free note-taking app, available at obsidian.md, works on Mac, Windows, and Linux
Your OpenClaw instance — running on a dedicated machine (a Mac Mini, spare laptop, or VPS)
A way to sync files between your machines (Obsidian's native sync at $5/month is the simplest option; a local network share or Tailscale also works)

Step 1: Set Up Your Obsidian Vault

Download and install Obsidian. When you create a new vault, you're essentially creating a folder of markdown files. Name it something simple without spaces — this matters later (spaces in the vault path can cause indexing issues).

Location matters: The vault needs to be accessible to both your personal machine and the machine running OpenClaw. Your options:

Obsidian Sync ($5/month) — simplest, works cross-platform, no network configuration needed
Shared network folder — works if both machines are on the same network; more friction when travelling
Tailscale — creates a persistent private network between your devices, so your "local" folder travels with you

For most people starting out, Obsidian's native sync is worth the five dollars to avoid the networking complexity.

Step 2: Build Your Vault Structure

An empty vault isn't useful. You need a folder structure that organises your information in a way your agent can navigate.

The best approach: ask your agent to build it for you.

Share a template of the folder structure you want (there are good starting templates available online for knowledge management vaults) and ask your agent to adapt it to your specific situation. Have it ask clarifying questions — what areas of your life do you want tracked? Work projects? Health? Family? Financial goals?

A typical vault structure might include:

/Inbox — a landing zone for raw notes and thoughts
/Projects — active projects with their own notes
/People — notes on relationships, contacts, colleagues
/Areas — ongoing responsibilities (health, finances, learning)
/Archive — completed projects and reference material
/Thinking — your ideas, frameworks, and mental models

The agent will create not just the folders, but also brief instruction files inside each one — explaining to future agents how that section should be used. This is the part that makes the system genuinely intelligent over time.

Step 3: Connect OpenClaw to the Vault

This step requires editing OpenClaw's configuration file (openclaw.json), located in the hidden .openclaw folder on your agent's machine.

Mac: Press Shift + Command + Period to show hidden files, then navigate to your user folder and find the OpenClaw directory.

Windows: Enable "Show hidden items" in File Explorer options.

Inside the config file, find the memory block and replace it with the Obsidian vault integration configuration. The key thing to update is the path — it should point to wherever your Obsidian vault is stored on the agent's machine.

Before making any changes: make a backup of the config file. A syntax error in a JSON file (a missing comma, an extra bracket) will prevent OpenClaw from starting. Keep a clean copy you can restore.

If editing JSON manually makes you nervous, just ask your agent to do it. Describe what you want — "connect my OpenClaw memory system to my Obsidian vault at [path]" — and it will handle the syntax correctly.

After saving, restart your agent to trigger indexing of the vault.

Step 4: Verify the Connection

Once your agent restarts, ask it to search your vault for a topic you know is in your notes. If it returns results, the integration is working.

A few things to test:

Search: "Search my vault for notes about [topic]"
Create: "Create a new note about [subject] and file it in the appropriate folder"
Analyse: "What patterns exist in my notes about [area]?"

That last one is where the system starts to feel genuinely different. An agent that can read through months of your notes and surface patterns, recurring themes, or gaps in your thinking is qualitatively more useful than one working from a compacted summary.

What This Enables

Once the integration is in place:

Persistent project memory. When you start a new conversation about a project, your agent already knows the history — decisions made, problems encountered, current status — because it's all in the vault.

Sharable notes. Files in the vault are accessible on your phone, other computers, or to other people you grant access. Your agent's output isn't trapped in a chat interface.

Structured personal knowledge base. The vault becomes a searchable record of how you think, what you're working on, and what you've learned — maintained and updated by your agent as you work together.

Better context for new conversations. Instead of re-explaining your situation every time you start a session, your agent can read the relevant vault sections and arrive already informed.

A Note on Privacy

One consideration worth flagging explicitly: if you're running OpenClaw with a cloud-hosted LLM (Claude, GPT, Gemini), the contents of your vault are being sent to that provider as context. Your notes, your projects, your personal details — all of it.

If you're using a local model via Ollama, this concern goes away. If you're using cloud models, be thoughtful about what you include in the vault. Sensitive financial information, private health data, and anything you'd be uncomfortable sharing with an AI provider should either be excluded or handled through a local-only model.

Getting the Most Out of It

The Obsidian integration is a foundation, not a finished system. The more you use it — the more notes you add, the more feedback you give your agent, the more structure you build — the more useful it becomes.

A useful starting habit: at the end of any meaningful work session, tell your agent to log a brief summary to the relevant project folder. Over time, this creates a rich record that makes future work significantly faster and more contextually aware.

Your agent's memory doesn't have to fade. With the right setup, it compounds.