The short answer
If you already use Obsidian and an LLM wiki, you have a place to keep what you know. Cognition adds a workflow for teaching agents how you want work done: capture a correction, approve a reusable skill, and recall it in a later session or another connected agent.
Cognition and Obsidian, side by side.
Swipe or scroll the table to compare both tools →
| What matters | Cognition | Obsidian |
|---|---|---|
| Primary job | Reuse approved working methods inside coding agents. | Write, connect, and organize a personal knowledge base. |
| What persists | Skills with steps, checks, reasoning, and the context where they apply. | Local Markdown notes, links, and attachments in your vault. |
| Getting it into an agent | Install Cognition in supported hosts; the bootstrap loads guidance and the agent calls cognition_assist. | Provide files or connect your chosen LLM workflow. Agent access depends on that setup. |
| Learning from a fix | The agent drafts a reusable skill; you review and approve it before saving. | Record the fix in a note. Automation depends on your plugins or scripts. |
| Across sessions and tools | Connected agents using the same brain can recall its approved skills. | Notes persist independently of chats; each agent needs a way to read them. |
| Where it lives | A hosted brain connected to your local coding agents. | Files on your device, with optional Sync and Publish services. |
One correction. The next session.
Imagine you have one hour after work, and your agent repeats a release mistake you already fixed.
With Obsidian
Save a deployment checklist in your wiki and have your LLM workflow read it before shipping.
With Cognition
Turn a deployment correction into an approved skill with triggers, steps, and checks. In a later relevant task, the connected agent can recall that procedure before acting.
Cognition makes approved guidance available through its installed instructions and tools. It does not guarantee that a model will follow every instruction. Review the result and keep consequential checks in your workflow.
Choose for the job you need done.
Choose Obsidian if…
Choose Obsidian for a local, editable knowledge base: research, linked notes, project context, and writing. Your existing LLM wiki may already be enough if you are happy maintaining how agents read and apply it.
Choose Cognition if…
Choose Cognition when the recurring problem is re-teaching an agent the same procedure or correction, especially when you switch between Codex, Claude Code, Cursor, and Gemini.
Can you use both?
Keep research and long-form context in your vault. When a working session reveals a repeatable procedure, ask your connected agent to draft a Cognition skill and approve the version you want reused. This is a complementary workflow, not a promise of automatic vault sync.
Common questions
Do I need to move my Obsidian vault?
No. Keep your notes where they are. Start by saving one approved procedure through a Cognition-connected agent; you do not need to migrate your knowledge base.
Is an AGENTS.md file enough?
For a small set of stable project rules, it may be. Cognition adds a shared skill workflow and task-relevant recall across connected agents. Its local bootstrap still uses the host’s instruction mechanism.
Sources and scope
Based on the official product pages linked below and Cognition’s documented skill workflow. This is not an exhaustive audit of plugins or custom integrations. Product capabilities can change.
Start with one fix worth remembering.
Connect your coding agent, capture a useful procedure, and approve what you want the next session to reuse.