Use these task-based guides to create, configure, and maintain Kiro bots. For an explanation of their architecture and behavior, see How bots work.
You can create a bot in two ways:
Ask any bot to create a new bot for you. What happens next depends on whether you include a goal in your request.
If you provide only a name, for example "Create a bot named Contributor," the bot shows a Create a bot guide card. Select Start to open the creation form with the name filled in, then complete the form as described in Option 2. Select Dismiss to cancel. The bot is created only when you confirm in the form.
If you provide a name and a goal, for example "Create a bot named Contributor with the goal of helping me contribute regularly to the Kiro Crew open-source project," the bot shows a New bot card. The card lists the name, the goal, any tools that run without asking, and the bot's reason for the request. To create the bot:
The card shows how long you have to respond.
The new bot's thread opens. It asks you to define a goal, or confirms the goal you entered during creation. Review the proposal card and approve it to continue.
You can change a bot's model in two ways:
Ask the bot to switch models. For example:
"Switch your model to [model name]."
The bot shows a Bot change card with the current model, the new model, and its reason for the change. Select Apply change to confirm, or Decline to keep the current model. The card also shows how long you have to respond.
By default, workers inherit the bot's model settings unless a separate worker model is configured. To use a separate model for delegated tasks, set the worker model and worker reasoning effort on the same page. This setting applies to new workers; existing workers retain the model and reasoning effort they started with.
A bot's permissions are set by its approval mode:
| Mode | Behavior |
|---|---|
| Normal | Requests approval for actions except specified built-in operations |
| Reads | Allows reads without approval; asks before other changes except specified built-in operations |
| Trust | Acts without per-action approval, subject to security policy |
New bots start in Trust, except bots imported by an app, which start in Normal. Changes apply to running workers. Security policy and blocked commands apply in every mode. See Permissions.
You can change a bot's permissions in two ways:
Ask the bot to change its permissions. For example:
"Change your permissions to Normal."
The bot shows a Bot change card with the current approval mode, the new mode, and its reason for the change. Select Apply change to confirm, or Decline to keep the current mode. Permissions don't change until you confirm. The card also shows how long you have to respond.
You can also begin under Goals → Create a goal → Tell <name> what to aim for. Check-ins are off by default. See Goals, milestones and check-ins.
The bot's standing thread and its worker sessions are separate. Stopping one worker is different from disabling Heartbeat or a custom schedule.
Use All schedules to review the wider schedule list. Heartbeat is configured separately in the first row of the bot's Schedules tab. Daily, weekly, and monthly goal check-ins remain under Goals. Agent schedules use credits when they call a model; choose timing that fits the task.
Use Fresh start when you want to continue with the same bot without replaying its current conversation context.
Fresh start stops a running turn and clears the active conversation and plan on the same thread. The saved transcript stays on disk; the next turn starts a fresh agent process. Saved memory, workspace files, and goals are not erased. Worker sessions and schedules are not all stopped by this action. Use their own controls if you want to stop background work.
If a worker needs a tool that is not available, it reports the requirement to its bot. The bot can request your approval through a card. You may be asked to enable the Dynamic capabilities preview before it can search for tools.
To add abilities manually:
The additions apply only to this bot. Workers reload them, after their current turn if busy. If the source agent includes instructions, those replace the bot's existing behavior directions. Abilities cannot override security policy.
You can always ask the bot directly in chat. For example: "Stop the worker that's running the tests" or "Turn off your scheduled check-ins."
To stop a bot or worker manually:
| To | Action |
|---|---|
| Stop a worker | Open Profile card → Sessions, select the worker, then Stop |
| Require approval before actions | Set permissions to Normal |
| Prevent scheduled wake-ups | Disable check-ins under Goals and Heartbeat under Schedules |
Deleting a bot unbinds it from new sessions. Its workspace, memory store, and past transcripts are kept. Clean Git worktrees are removed; worktrees with uncommitted changes and their branches remain. If you delete the built-in bot, Kiro Crew does not recreate it automatically.
| Problem | Resolution |
|---|---|
| The bot does not answer | Run kirocrew status; if needed, start the gateway with kirocrew gateway, then run kirocrew doctor |
| A worker appears stuck | Open Profile card → Sessions to inspect or stop it |
| An approval card disappeared | Cards expire if not acted on; ask the bot to propose the change again |
| A worker lacks a tool | Add abilities to the bot's workers |
Manage bots