Three layers of the knowledge base

Cadau splits materials that agents read into three layers so you write the right content in the right scope. Usage: Using the knowledge base.

Cadau’s interface can be Chinese or English. The help agent answers in your interface language and reads the matching system knowledge tree (help/ or help/en/).

System layer (global)

• Who uses it: the help agent (no workspace selected, or built-in help chosen in Messages), and any scene that needs one platform-wide explanation.
• What to put: new-user guide, navigation, account and security, product overview (this help/ / help/en/ tree).
• Where to maintain: platform admins edit in Admin console → System knowledge base (deploy directory defaults to help/; English UI prefers help/en/ when present).
• After edits: in the admin console run Rebuild index with AI to refresh the root and each theme’s index.json.

Workspace layer (team)

• Who uses it: every member’s agents in the same workspace consult it in chat.
• What to put: team conventions, project background, shared processes, internal wiki.
• Where to maintain: after sign-in, Workspace collaboration → Manage workspace knowledge base (that workspace must be selected).
• After edits: on the knowledge editor, Generate index with AI (use Force regenerate if the body did not change).
• Bulk migrate: the knowledge editor can upload a folder / archive or download an archive (current browse location; .md and index.json only).
• Source archive: the same page can switch to Source archive to upload historical Word / Excel / PPT / PDF / images; mark current or archived. Put old policy and old notices here — do not rewrite them as how-to articles.

Personal layer (My agents)

• Who uses it: only that agent instance in chat; by default not shared with other members in the same workspace.
• What to put:

- Knowledge base: role setup, dedicated wording, private materials, business metrics as Markdown; - Memory (separate from the knowledge base): your preferences, confirmed decisions, project milestones — personal material across conversations (see Agent memory).

• Where to maintain:

- Knowledge: Agents → My agents → Knowledge base; - Memory: the same instance’s Memory page, or Remember in Messages.

• After edits: the knowledge base needs Generate index with AI on the editor; memory entries take effect on the next round (auto-extract runs asynchronously after each Q&A).
• Bulk migrate: the agent Knowledge base editor also supports upload / download archive (same rules as the workspace layer).
• Source archive: same page; historical files for this agent only; current/archived meaning matches the workspace layer.
• Runtime file roles (persona, operating rules, profile, memory): see Practice guides.

How chat uses this

The agent first pulls relevant passages (how-to articles and source archives share one retrieval). Under the answer, sources list titles or original file names; open to preview. After you edit an article or ingest an original, the next round can use the new content. Index index.json still boosts article selection — regenerate the index after the body changes so summaries are not stale.

Ordinary Q&A uses current archives by default; questions about “before / old notices” look at archived items.

Load only for a scene (optional): an index entry may include paths (e.g. ["/*.tsx"]). That article enters chat only when the user message or attachment path matches. Do not fill this** on general intros and account notes. For system knowledge in the admin console, see System knowledge base and retrieval index; workspace and agent knowledge support the same — generate the index on the matching editor.

Relation to the action assistant

On Agents, Skills, Workspace collaboration, Contacts, Account, and similar pages, the action assistant also answers with the current module’s context. Its knowledge still comes mainly from workspace and system articles; the personal layer follows the bound agent.

The help agent (in Messages, especially with no workspace selected or right after registration) mainly reads system-layer docs in this tree. Team and personal layers are maintained in Workspace collaboration and My agents.