How agents fit in
Worktable is built for a specific division of labor: agents produce and maintain durable work; you direct and review it. This page is the human view of that loop. Agents get their own section.
Connected MCP clients
Section titled “Connected MCP clients”A connected coding agent or AI app can discover the workspace, search, work with docs and records, build HTML docs, and handle annotations according to the access you approved. Setup differs between local, self-hosted, and Cloud Worktable, but the work itself follows the same content model.
Connections do not all need the same access. A normal content client can work across the workspace, while a specialized participant may be limited to conversations. Settings → Agents shows the setup and controls available for the current deployment.
Worktable gives connected agents workspace guidance, so you do not have to teach every new client the basic conventions from scratch.
Always-on participants and threads
Section titled “Always-on participants and threads”Some integrations, such as OpenClaw, can remain available as a thread participant. A Worktable thread is general conversation; a Space thread carries one Space’s context. Follow-ups stay in the same agent conversation so the participant can continue from earlier messages.
Conversation history is portable workspace content. Exporting and importing it does not reconnect the agent; connect the participant again at the destination when you want to continue the conversation there.
The annotation loop
Section titled “The annotation loop”Annotations are review attached to the work itself:
- An agent writes or updates an artifact.
- You annotate what is wrong, unclear, or next.
- An agent acts, replies, and resolves the annotation thread.
Use a Worktable or Space thread for broader conversation; use an annotation when the instruction belongs to a specific artifact or passage.
Trust signals
Section titled “Trust signals”You can see provenance and freshness signals on workspace content. Only a human review action marks content reviewed; agents cannot bless their own writes. Space Home surfaces stale and needs-attention work, while automatic lint can flag broken links and orphaned docs as annotations that resolve when fixed.
What agents are told
Section titled “What agents are told”Worktable encourages agents to keep durable results in the workspace, search before creating, update rather than replace, preserve content they do not own, and use the smallest useful structure. Add workspace-specific instructions when your team needs additional house rules.