Control

Tools and approvals

Minds can use tools to look things up, create views, update files, or take other actions. Approval prompts keep you in charge before sensitive work happens.

What a tool call means

A tool call is a request from a mind to do something outside a normal chat reply. It may read a file, search a source, create a Lens view, open a Canvas dashboard, schedule work, talk to another mind through A2A, or perform another action Chamber supports.

  • Reading tools gather information so the mind can answer better.
  • Writing tools may create, update, or delete something.
  • External tools may contact a service or use an account you connected.
  • Scheduled tools can run later, even when you are not watching the chat.

Built-in Chamber tools

Chamber gives minds several built-in tool groups. Marketplace tools can add more, but these are the core Chamber-managed tools users are most likely to see.

A2A collaboration a2a_list_agents, a2a_send_message, a2a_send_task, a2a_get_task, a2a_list_tasks, and a2a_cancel_task let minds discover each other, exchange messages, delegate tracked tasks, check status, and stop in-flight work.
Canvas canvas_show, canvas_update, canvas_close, and canvas_list create, refresh, close, and list browser-based HTML canvases for dashboards, reports, forms, and rich visual output.
Scheduled work cron_create, cron_list, cron_remove, cron_enable, cron_disable, cron_run_now, and cron_history create and manage recurring prompts, process jobs, webhooks, notifications, and run history.
Lens views Lens uses validated view.json files to render forms, tables, briefings, detail views, status boards, timelines, editors, and canvas-backed views inside Chamber.
Marketplace CLI tools Marketplace sources can install npm-global or GitHub release asset tools. Chamber advertises each installed tool to minds with its command name, description, help command, and agent instructions.

Tool name directory

a2a_list_agents Finds other available minds and their skills before delegating work.
a2a_send_message Sends a quick peer message to another mind.
a2a_send_task Creates a tracked asynchronous task for another mind.
a2a_get_task Checks one task's state, history, artifacts, and result.
a2a_list_tasks Lists tasks, optionally by shared context or status.
a2a_cancel_task Cancels an in-progress A2A task.
canvas_show Creates or opens a browser canvas from HTML content or an existing HTML file.
canvas_update Updates an existing canvas and live-reloads the browser page.
canvas_close Closes one canvas or all canvases for a mind.
canvas_list Lists currently open canvases for the active mind.
cron_create Creates a scheduled prompt, process, webhook, or notification job.
cron_list Shows scheduled jobs with next run and recent run details.
cron_remove Deletes a scheduled job and its stored run history.
cron_enable Turns a paused job back on.
cron_disable Pauses a job without deleting it.
cron_run_now Runs a scheduled job immediately and records the result.
cron_history Shows recent run history, optionally for one job.

How approval prompts work

When Chamber needs your decision, it shows an approval prompt with the action, the mind requesting it, and the details Chamber can show safely. Read it before choosing.

1

Check the action

Look for plain signs of what will happen, such as read, write, send, schedule, or open.

2

Check the target

Make sure the file, folder, account, link, or service is the one you expected.

3

Choose approve or deny

Approve if it matches your request. Deny if it is unclear, too broad, or surprising.

Safe choices

You can always ask the mind to explain the action in simpler words before you approve. Denying a prompt does not break Chamber. It only stops that action.

  • Prefer small, specific actions over broad access.
  • Do not approve access to private information unless the task truly needs it.
  • Pause if the prompt mentions an account, file, or website you did not name.
  • Use deny when a request feels rushed, vague, or unrelated to your goal.