Dispatch·

hand a task to another persona (voice/telegram/thinker) through the shared brain — tools/dispatch.py, 1 tool.

From the module docstring

Dispatch — spawn a devduck agent IN-PROCESS as a tool, optionally on a schedule.

tool	does
`dispatch`	Spawn a devduck instance as a tool, optionally on a schedule.

`dispatch`·

dispatch(
    action: str = 'run',
    prompt: Optional[str] = None,
    mode: str = 'sync',
    model: Optional[str] = None,
    model_provider: Optional[str] = None,
    tools: Optional[str] = None,
    system_prompt: Optional[str] = None,
    mcp_servers: Optional[str] = None,
    env_vars: Optional[Dict[str, str]] = None,
    work_dir: Optional[str] = None,
    timeout: int = 600,
    dispatch_id: Optional[str] = None,
    enable_servers: bool = False,
    load_tools_from_directory: bool = True,
    tail: int = 100,
    schedule: Optional[str] = None,
    run_at: Optional[str] = None,
    runs: Optional[int] = None,
    schedule_id: Optional[str] = None,
) -> Dict[str, Any]

Spawn a devduck instance as a tool, optionally on a schedule.

Builds an in-process devduck.DevDuck agent with custom config, runs the prompt, persists agent.messages to SQLite, writes a summary to the unified cross-persona log, and (optionally) schedules recurring fires.

argument	meaning
`prompt`	Query for the dispatched agent (required for run/schedule)
`model`	STRANDS_MODEL_ID override
`model_provider`	bedrock / anthropic / openai / ollama / etc
`tools`	DEVDUCK_TOOLS string
`system_prompt`	Custom SYSTEM_PROMPT for the child
`mcp_servers`	MCP_SERVERS JSON string
`env_vars`	Dict of extra env vars
`work_dir`	cwd for the child
`timeout`	sync-mode timeout (seconds, default 600)
`enable_servers`	allow child to start its own zenoh/proxy/etc
`load_tools_from_directory`	load ./tools/*.py auto-discovery (default True)

Actions

Run/inspect: - "run": spawn + run NOW. Modes: sync | async | bg. - "list": list dispatch runs - "status": inspect one (dispatch_id required) - "result": final assistant text (dispatch_id required) - "messages": full agent.messages dump (dispatch_id required) - "logs": tail captured stdout (dispatch_id required) - "wait": block until done (dispatch_id required) - "clean": purge finished runs older than 24h Schedule: - "schedule": register a schedule (requires prompt + schedule|run_at) - "schedules": list active schedules - "unschedule": remove schedule (schedule_id required)

Modes (action="run"): sync — block + return text async — thread-spawn, return id bg — alias for async

Scheduling: schedule="*/5 * * * *" → cron expression, fires on match run_at="2026-05-26T18:00" → one-shot at ISO datetime runs=N → cap total fires (default ∞ for cron, 1 for run_at)

Scheduled fires use the same model/tools/system_prompt/etc you pass in this call. Ticker runs every 30s. Schedule survives only while THIS process is alive (in-process; if you need cross-process persistence, use devduck's own scheduler tool).

Returns

Dict with status + content.

source: tools/dispatch.py:308

Dispatch·

dispatch·

`dispatch`·