OpenClaw + Obsidian: Build an AI-Powered Second Brain Agent

The mistake most Obsidian users make is treating the vault as a passive archive. Notes go in, rarely come out. OpenClaw changes that dynamic completely. Your agent can query the vault for relevant context before answering a question, append new insights to existing notes, and create structured summaries of external content — all without you opening Obsidian at all.

Why OpenClaw + Obsidian Works So Well

Obsidian's plain-text Markdown format is what makes this integration exceptional. There is no database to reverse-engineer, no proprietary format to decode. Every note is just a .md file that OpenClaw can read and write directly through the filesystem.

The result is a genuinely personal AI. Instead of a generic assistant drawing on web knowledge, you get an agent that knows your specific projects, your previous decisions, your meeting notes, and your research findings. That context gap is what separates useful AI from impressive-but-generic AI.

Here's what we've seen consistently in production deployments: users who connect OpenClaw to their vault report 40 to 60 percent fewer "I know I wrote this down somewhere" searches. The agent surfaces it before they even ask.

Installing the Obsidian Skill

Install via the OpenClaw CLI or through ClaWHub. The skill has no external dependencies beyond your vault path.

# Install the skill
openclaw skill install openclaw-obsidian-skill

# Confirm installation
openclaw skill list | grep obsidian
# openclaw-obsidian-skill  v1.8.2  installed  inactive

No OAuth flow is required for filesystem mode. You simply point the skill at your vault directory. API mode requires the Obsidian Local REST API community plugin.

Vault Access Modes

Filesystem Mode

Filesystem mode is the default and simplest option. OpenClaw reads and writes .md files directly. This works whenever OpenClaw runs on the same machine as your Obsidian vault — which covers most personal setups.

The skill respects your vault's folder structure exactly. Notes live at their real paths, links are standard Obsidian wiki-link format, and frontmatter is parsed from the YAML block at the top of each file.

REST API Mode

REST API mode uses the Obsidian Local REST API community plugin. Install it from the Obsidian community plugins browser, enable it, and note the API key and port it generates.

Use API mode when OpenClaw runs on a different machine than your vault, or when you want finer-grained control over what the agent can access. The API surfaces vault operations as HTTP endpoints that OpenClaw calls directly.

Configuration Reference

Here is a complete working configuration for the Obsidian skill in filesystem mode, covering the most common production setup.

# ~/.openclaw/skills/obsidian.yaml
skill: openclaw-obsidian-skill
version: "1.8"

vault:
  mode: filesystem              # filesystem | api
  path: "~/Documents/MyVault"   # absolute path to vault root
  # For API mode, use instead:
  # mode: api
  # api_url: "http://localhost:27123"
  # api_key: "${OC_SECRETS}/obsidian_api_key"

access:
  allowed_actions:
    - read_note
    - search_vault
    - append_note
    - create_note
  # Uncomment to also allow deletion (use with caution)
  # - delete_note

  excluded_paths:
    - "Templates/"
    - "Archive/"
    - "_MOCs/"
    - "Attachments/"

  write_mode: append_only        # append_only | full_write
  max_note_size_kb: 512          # skip notes larger than this

notes:
  required_frontmatter_fields:
    - tags
    - created
    - status
  default_frontmatter:
    status: "ai-generated"
    created: "{{date}}"
  output_folder: "AI-Notes/"    # where new notes go
  link_format: wiki              # wiki | markdown

The excluded_paths list is your safety net. Add any folder you never want the agent to touch. Templates and MOCs especially should be off-limits for write operations.

Workflow Examples

Automatic Meeting Notes

After a call, trigger the agent with a brief voice memo or typed summary. OpenClaw reads the meeting context from your vault (previous meeting notes, project status, attendee notes), then creates a structured meeting note with action items, decisions, and follow-up tasks — already linked to the relevant project notes.

Research Capture Pipeline

Pair the Obsidian skill with a web reading workflow. When you save a URL to a reading list, OpenClaw fetches the article, extracts key insights, and appends them to a running research note in your vault — complete with the source URL and a brief summary. Your vault grows automatically from your reading habits.

Context Retrieval for Answers

This is the sleeper use case. Before the agent answers any complex question, configure it to run a vault search first. The agent surfaces relevant notes, injects them as context, and answers using your own prior research. The difference in answer quality is immediately noticeable.

We'll get to the exact prompt template structure in a moment — but first you need to understand why the search_vault action quality depends entirely on your vault's tagging consistency.

Daily Note Generation

Schedule the agent to run each morning. It reads yesterday's daily note, checks your calendar for today's events, reviews open tasks across your project notes, and creates today's daily note pre-populated with the right context. Opening Obsidian at 9am means your workspace is already set up.

Common Mistakes

Here's where most people stop and wonder why the skill isn't working as expected.

• Not setting excluded_paths — the agent will happily read and potentially overwrite your templates and MOCs if you do not exclude them explicitly
• Using full_write mode immediately — start with append_only to prevent the agent from replacing note content instead of adding to it
• Missing frontmatter on existing notes — the skill's search relies heavily on frontmatter tags; vault notes without consistent tagging return poor search results
• Setting output_folder to the vault root — AI-generated notes should go in a dedicated subfolder so they are easy to review and the vault structure stays clean
• Not testing search quality before enabling writes — run search_vault queries in dry-run mode for a week to confirm the agent retrieves relevant notes before you let it write anything

Frequently Asked Questions

Does the OpenClaw Obsidian skill work with vaults stored in iCloud or Dropbox?

Yes, as long as OpenClaw can access the vault path on your filesystem. iCloud Drive and Dropbox sync the vault locally, so the skill reads and writes the local copy. Sync conflicts are rare but can occur if you edit a note while the agent is also writing to it simultaneously.

Can OpenClaw create new notes, or only read existing ones?

Both. The skill supports read, write, append, and delete operations. Most users enable read and append first, then enable write for new note creation once they trust the output format and frontmatter structure produced by their prompt templates.

Will the skill break my Obsidian graph view with auto-created notes?

Only if your prompt templates create unlinked notes. Structure your templates to include backlinks to relevant MOCs or index notes. The graph view reflects what is in your vault — the skill writes what you instruct it to write through your prompt configuration.

How does OpenClaw handle Obsidian's YAML frontmatter?

The skill parses and writes standard YAML frontmatter. Configure required_frontmatter_fields in the skill config to ensure every AI-generated note includes your tags, created date, and status fields automatically with every write operation.

Can I use the Obsidian skill without the Obsidian Local REST API plugin?

Yes. The skill has two modes: direct filesystem access (reads and writes .md files directly) and API mode via the Local REST API plugin. Filesystem mode works on the same machine as your vault. API mode enables remote vault access across devices.

What is the best way to prevent OpenClaw from overwriting important notes?

Set write_mode: append_only for notes in your protected folders. Use the excluded_paths config to block the skill from touching your templates, MOCs, or archive folders entirely. Test with a dedicated AI-output folder first before widening write access.

Does the skill support Dataview queries for knowledge retrieval?

Not natively, but the skill can read note frontmatter and pass it to the agent context. For Dataview-style queries, use the skill's search_vault action with tag or property filters, which produces equivalent results for most retrieval use cases.

You now have a working foundation for an AI second brain. The Obsidian skill is live, vault access is configured, and you understand how to keep your important notes safe while the agent grows your knowledge base automatically.

Start with search and read. Add append after a week. Enable create_note once your output folder has notes you are genuinely happy with. That staged approach is how you build a system you trust rather than one you babysit.