Choose an assistant conversation, a saved @ mention, or an external MCP client. Each has a different execution and review path.
Choose a workflow
| Workflow | How it starts | How changes are saved |
|---|---|---|
| Tano assistant | Send a question or instruction in the sidebar or inline | Explicit edit instructions can execute; requests for proposals stay pending |
| @Agent task | Select a configured agent in a document, annotation, or comment, then save or post | The task follows the request and records tool receipts |
| MCP Gateway | Connect an external client to a local workspace | Default review access submits pending proposals |
| External file tools / MCP write access | Explicitly authorize direct file operations | Direct writes; Tano detects external changes and protects conflicts |
Mention an agent
Configure a model connection, then check its agent identity, model, and automatic-processing setting in the agent inbox. Type @, select the agent from the picker, and write a specific task in the same context. Save the document or publish the comment to dispatch it.
Agent mentions inserted by the picker carry occurrence metadata to deduplicate delivery. Handwritten [[object://person/ID]] references without this metadata, code samples, and ordinary person references do not dispatch tasks. Renaming an agent does not change its identity.
The inbox shows queued, running, waiting for a reply, completed, failed, cancelled, or uncertain states. Open the source and execution record to inspect results. If the agent asks a question, reply to that message in the associated discussion. Cancellation stops further work but does not undo saved changes. Check uncertain write results before retrying; resending a missing reply is different from rerunning file edits.
Local and hosted execution
Local agent tasks use this device's connections and require the app to remain running. Closing an assistant panel does not stop the app's task runner; closing the app or switching away from its workspace is a different lifecycle boundary. The demo uses simulated responses only.
Official online workspaces can use server-managed agents when the service has enabled models and task execution. Those queues and sessions are stored by the service and can run independently of a browser tab. This does not make your local Codex or Claude CLI available to the server. Availability depends on the connected deployment, and hosted actions remain subject to the requesting user's current access.
Share a local working folder
Open a folder in desktop Tano and grant your external agent the intended directory and tools. Its Markdown changes appear in Tano. For narrower access or reviewable batches across documents, objects, and tables, use MCP. Write shared conventions in workspace guidance.
Tano compares external changes with your baseline, merges independent edits, and preserves overlapping changes for review. Keep important history in Git or a backup. A person's name or an agent's self-reported label is not proof of authenticated identity.