Collect ideas and turn them into knowledge or skills

Collect pages and ideas now, and decide later what becomes knowledge or a skill. Nothing you collect is knowledge until a signed-in person confirms it in the Cockpit.

The path is always the same: collect → review in chat → confirm → knowledge or skill.

What you need #

• For web pages: the Web Clipper in Chrome, installed and connected. To install, update or troubleshoot the Web Clipper, see "Collect web pages with the Web Clipper".
• To collect: an account with intake:read and intake:write.
• To confirm: a browser session in the Cockpit with intake:confirm, plus knowledge:write for knowledge, or skill:write and skill:activate for a skill. Replacing skills also needs skill:archive for each replaced skill.

Collect something #

1. Open the page.
2. Optional: select the text you want as an excerpt.
3. In the toolbar, select the Web Clipper icon ("SubKit Clipper").
4. Check the preview. The source address is captured automatically.
5. Optional: add why it matters.
6. Pick a category. Your default category is preselected, at first inbox.
7. Select Collect.

Learnings from sessions with AI assistants arrive in the same Inbox. Their origin names the assistant with its mark, such as Claude Code, Codex or Chat. When the assistant is unknown, it reads AI assistant. Collected pages carry the origin Browser.

Add a note #

A note is an item without a source address, such as a thought or an idea.

1. In the Cockpit, open Inbox and select Add entry. If the Inbox is still empty, select Add a note.
2. In the Add to intake dialog, enter a Title.
3. Write the text under Content. Markdown is allowed.
4. Pick a Category, or create one with New category….
5. Select Add.

The note carries the origin Note. Its page shows no Open original page link, because there is no source.

Review and organize items #

1. In the Cockpit, open Inbox. (opens in a new tab)Screenshot from build 10.6.0
2. Optional: in the line above the list, click the hint, for example "3 entries from your AI assistants are waiting for your review". You then only see items from your AI assistants. The filter Origin: AI assistant appears next to the search; click it to clear it.
3. Click an item. On the Content tab, one sentence names the origin, such as "Collected by Chat in the session … · for … · …". On the right, Open conversation opens the chat conversation; for collected pages, Open original page opens the source.
4. Optional: select Edit item to change title, content or category. The captured original is preserved.
5. Optional: Discard or Restore.

Inbox visibility follows the permissions of your workspace. Categories are not private.

Review in chat #

Review one item:

1. In the Inbox, open the item.
2. Select Review in chat. The chat shows the item as an @ reference.

Review a whole category:

• In a new chat, start the category's review suggestion, or ask: "Look through my new Research entries and suggest what to keep."

The chat prepares a proposal with content, sources and destination, and links to it. The link opens the item's Review tab. Completed decisions appear on the History tab under Proposals and decisions.

Confirm a proposal #

You need a browser session in the Cockpit, intake:confirm and the destination right. API keys, MCP clients and the Web Clipper cannot confirm.

1. Open the item and select Review.
2. Review the proposal: content, sources and destination.
3. Select Confirm as source or Confirm as skill.

Only then can ordinary knowledge retrieval use the result, or agent briefings carry the skill. For a skill, confirming also archives the skills it replaces. The proposal shows each change.

Good to know:

• Moving, editing or discarding an item never makes it knowledge.
• If a proposal or its sources change, you review it again.
• SubKit replaces credentials such as API keys and tokens before saving an item.
• The item's original stays linked to the knowledge or skill.
• You keep maintaining confirmed sources and skills as described in "Maintain sources, skills and files".

If something does not work #