Tools¶

Generated from strands_robots/tools/*.py at build.

Every tool is a plain function first. Mutating tools take dry_run=True, default to mode="sim", and refuse real motion without confirm=True; on top sits the strands HITL gate for anything that moves hardware. The list is strands_robots.tools.STRANDS_ROBOTS_TOOLS.

14 tools in 6 modules.

tool	module	parameters	does
`camera_list`	`camera_tools`	`max_index`: `int` = `MAX_INDEX`	Probe camera indices `0..max_index` and list the ones that open, with their frame size.
`camera_snap`	`camera_tools`	`index`: `int` = `0` `path`: `str \| None` = `None`	Grab one frame from camera `index` and save it as PNG (default under `$STRANDS_ROBOTS_HOME/snaps`).
`mesh_fleet`	`mesh_tools`	`wait_s`: `float` = `2.0` `backend`: `str \| None` = `None`	Listen on the mesh for `wait_s` seconds and list every peer heard: robot, liveness, state age, last state.
`mesh_ask`	`mesh_tools`	`peer`: `str` `verb`: `str` `args`: `dict[str, Any] \| None` = `None` `timeout_s`: `float` = `5.0` `backend`: `str \| None` = `None` `dry_run`: `bool` = `True` `confirm`: `bool` = `False`	Send one verb to a mesh peer and return its reply. Motion verbs (`act`, `execute`) pass the consent rail.
`mhs_fleet`	`mhs_tools`	`broker`: `str \| None` = `None` `allow_insecure`: `bool \| None` = `None`	Discover every device on the MHS broker: id, type, labels, status, procedures and state keys.
`mhs_invoke`	`mhs_tools`	`device`: `str` `procedure`: `str` `args`: `dict[str, Any] \| None` = `None` `broker`: `str \| None` = `None` `dry_run`: `bool` = `True` `confirm`: `bool` = `False` `allow_insecure`: `bool \| None` = `None` `timeout_s`: `float` = `30.0`	Invoke one procedure on one device. Motion (`act`, `execute_op`) passes the consent rail; `stop` always goes.
`list_policies`	`policy_tools`	-	List the policy providers: name, the extra that installs it, the frame it emits, whether it needs images.
`run_policy`	`policy_tools`	`name`: `str` `policy`: `str` = `'mock'` `instruction`: `str` = `'hold still'` `max_steps`: `int` = `100` `mode`: `str` = `'sim'` `port`: `str \| None` = `None` `control_hz`: `float` = `50.0` `replan_ratio`: `float` = `0.5` `dry_run`: `bool` = `True` `confirm`: `bool` = `False` `record`: `str \| None` = `None`	Roll out a policy on a robot and report the steps, why it stopped and the final joints.
`record_episode`	`record_tools`	`name`: `str` `repo_id`: `str` `policy`: `str` = `'mock'` `instruction`: `str` = `'demonstration'` `max_steps`: `int` = `300` `mode`: `str` = `'sim'` `port`: `str \| None` = `None` `leader`: `str \| None` = `None` `leader_port`: `str \| None` = `None` `fps`: `int` = `30` `root`: `str \| None` = `None` `dry_run`: `bool` = `True` `confirm`: `bool` = `False`	Record one episode into a lerobot dataset: a policy rollout, or a leader-arm teleop when `leader` is set.
`dataset_info`	`record_tools`	`path`: `str`	Episodes, frames, fps, robot type and feature names of a lerobot dataset directory.
`list_robots`	`robot_tools`	-	List every robot in the registry with its joints and the modes (sim, real, twin) it can open.
`observe`	`robot_tools`	`name`: `str` `mode`: `str` = `'sim'` `port`: `str \| None` = `None`	Read one observation from a robot: joints in its frame, camera names, scalar sensors. Never moves it.
`move`	`robot_tools`	`name`: `str` `action`: `Action` `mode`: `str` = `'sim'` `port`: `str \| None` = `None` `dry_run`: `bool` = `True` `confirm`: `bool` = `False` `settle_s`: `float` = `1.0`	Move joints to targets and hold until they settle; report the commanded, observed and remaining error.
`preflight`	`robot_tools`	`name`: `str` `policy`: `str` `mode`: `str` = `'sim'` `port`: `str \| None` = `None`	Check that `policy` can drive `name`: frame and DOF match and the policy's inputs are observed.

from strands import Agent
from strands_robots.tools import STRANDS_ROBOTS_TOOLS

agent = Agent(tools=STRANDS_ROBOTS_TOOLS)
agent("observe so101 in simulation")