Automate tasks with Botize Agents

Actions

The action is what happens automatically, without you having to do anything else.

• 

  Invoke agent

  Invoke an agent by its key. The step is just a pointer: the agent's identity (role, brain, tools) lives in the config.json of its folder, next to its documents and memory.

• 

  Compose think (text-driven brain turn)

  Internal activity: renders one brain turn as prompt + JSON Schema for text-driven brains (claude-code bridge). The loop yields claude_code/run with the result.

• 

  List connections (builder)

  Internal builder tool: lists the account's connections by app (names only).

• 

  Search methods (builder)

  Internal builder tool: text search over the node's app catalog, with each method's real form fields.

• 

  Write agent config (builder)

  Internal builder tool: validates and writes a new agent's config.json (the agent's identity and security boundary).

• 

  View agent (builder)

  Internal builder tool: list the account's agents or read one agent's config.json.

• 

  Create a Botize task (meta-tool)

  Create a Botize task from a SIMPLE definition: {description, steps:[{app, method, connection, fields, rename}]}. Step 0 must be a trigger (app 'type' each|request|instant, or a webhook_* method). Field values may reference previous steps' output vars as {{var}}; 'rename' renames a step's own output vars to avoid shadowing. Validation is strict against the real node (methods, connections, fields, vars) and errors come back as text to fix and retry. The task is created PAUSED: the user activates it from the Botize panel.

• 

  Run one of the user's Botize tasks (task-as-tool)

  Expose one of the account's Botize tasks as an agent tool: the agent launches it with input values and receives the task's final output when it ends — even if that takes hours (e.g. a human confirmation step). Bind with task_id FIXED (the task is pinned, never chosen by the model) and expose the task's declared input vars. The task must be ACTIVE and runs deterministically: whatever guarantees the user wired into it cannot be skipped by the agent.

• 

  Get connection docs

  Internal activity: fetch the official connection documentation of a Botize app (guide, errors, FAQs) in one call, from api/botize-docs. Used as a tool by connection-assistant agents.

• 

  Knowledge base

  Internal activity: query the Botize knowledge base (the same curated docs tree the claude.ai MCP reads) under api/botize-docs. mode 'search' greps across docs; mode 'read' cats one doc. Used as a tool by operator/builder agents to learn the real Botize shape of things (conditions, step fields, formulas).

• 

  Read a Botize task's definition

  Return the full definition (trigger + steps + their parameters and {{vars}}) of one of the account's Botize tasks, so the agent can inspect it, explain it, or prepare an edit. Read-only. Pass the numeric task_id.

• 

  List the account's Botize tasks

  Return the account's Botize tasks as a list of {task_id, description, steps}, so the agent can show a menu or resolve which task the user means. Read-only, no parameters.

• 

  Inspect a task's recent runs (and why it failed)

  Return diagnostics about a task's recent executions: whether it ran, its status, and if it is failing, the reason. Read-only introspection the panel does not expose conversationally. Pass the numeric task_id.

• 

  Edit and save a Botize task

  Full-replace a task's definition with a modified microbot array (the SAME shape read_task returns under 'task'). Flow: read_task first, modify the array, then update_task with the FULL modified array. Validated against the real node; errors return as text to fix and retry. Preserves the task's active/paused state.

• 

  Reschedule a task's next run

  Change the NEXT execution date/time of an ACTIVE scheduled task. Pass task_id and 'when' (absolute date-time or unix timestamp, at least 60s ahead). Only works on active interval tasks; a paused task returns an error.

• 

  Activate a paused task

  Activate (start) a paused task so it attends its trigger/schedule again. Pass task_id.

• 

  Pause an active task

  Pause (stop) an active task so it stops attending its trigger/schedule until reactivated. Pass task_id.

• 

  Ask another agent (agent-as-tool)

  Expose another agent of this account as a tool: the caller delegates a question and receives the delegate's final answer. One-shot: clean question, answer, forgotten (the delegate's durable memory still applies). Bind with agent_key FIXED (the delegate is pinned, never chosen by the model) and question EXPOSED (the model writes it).

• 

  Think (agent brain turn)

  Internal activity: one brain turn for the agent loop. Given the conversation and the available tools, returns a final text or a tool call.

• 

  Run tool (agent tool call)

  Internal activity: run one of the agent's tools (grep/sed/...) jailed to the agent's folder, and return the result.

• 

  Load conversation (agent short-term memory)

  Internal activity: load the agent's prior conversation for a session, to continue the dialogue.

• 

  Save conversation (agent short-term memory)

  Internal activity: persist the agent's conversation for a session, so the next episode continues the dialogue.

• 

  Meter brain usage (agent leash)

  Internal activity: charge a claude-code turn's token usage to the agent's daily budget.

• 

  Budget gate (agent leash)

  Internal activity: admit or politely deny a message against the agent's daily token budget.