Skip to main content

steadyhand

Turn a written lab protocol into a robot program, and rehearse it before it touches a real bench.

tests PyPI Python versions License: Apache-2.0 Status: pre-alpha

An 8-channel head draws buffer from a reservoir and fills a 96-well plate column by column on a Hamilton STARlet deck, in the physics scene the rehearsal runs in

Write what you want done in a short protocol file that names no robot, or have a language model draft that file from your notes or a kit manual. steadyhand checks that the liquid adds up, plans the run for the robot you pick, replays every move in a MuJoCo scene built from the same deck, and writes the program that does exactly what was rehearsed.

notes, then a protocol file, checked, planned, rehearsed, and the robot program

Pre-alpha. The whole path runs for pipetting protocols on a Hamilton STARlet, an Opentrons Flex and an Opentrons OT-2, but nothing has been compared with a real robot yet. It is not a tool to trust with a real run. See what it does not do yet.

Hamilton STARlet Opentrons Flex Opentrons OT-2
Program it writes PyLabRobot script Opentrons Python protocol Opentrons Python protocol
Physics replay of every move yes yes yes
The maker's own simulator yes yes, a release that still takes OT-2 files
Magnet for bead cleanups magnetic plate adapter Magnetic Block Magnetic Module
Planted mistakes stopped 24 of 26 18 of 21 18 of 21

Try it

pip install steadyhand
steadyhand demo

demo rehearses a small protocol that ships inside the package. To run the examples in this repository:

git clone https://github.com/dundysm/steadyhand && cd steadyhand
pip install -e .
steadyhand rehearse examples/serial_dilution.yaml

steadyhand rehearse prints three verdicts: check PASS with what each liquid uses, plan OK with the deck layout and tips, and rehearsal PASS with 1088 steps, no collisions and no tips out of the liquid

steadyhand-out/ then holds protocol.py (the PyLabRobot script), replay.html (the run played back, below), scene.xml (open it with python -m mujoco.viewer --mjcf steadyhand-out/scene.xml), a report as text and JSON, and two pictures when the machine can render offscreen. Every command exits with status 1 when something is wrong, so it can gate a script or a CI job.

The protocol file

The protocol in the run above is this file:

steadyhand: 0.1
name: Two-fold serial dilution across a 96-well plate
labware:
  plate:     {type: Cor_96_wellplate_360ul_Fb}
  reservoir: {type: nest_1_troughplate_195000uL_Vb}
liquids:
  buffer: {in: reservoir/A1, volume: 20 mL}
  sample: {in: plate/A1:H1, volume: 200 uL}
steps:
  - transfer: {from: reservoir/A1, to: plate/A2:H12, volume: 100 uL, tips: reuse}
  - serial_dilution:
      start: plate/A1:H1
      steps: 11
      direction: right
      volume: 100 uL
      mix: {times: 3, volume: 80 uL}

It says what to do and nothing about how a robot does it. Every amount carries a unit, unknown fields are errors, and every problem is reported at once with its place in the file. The format is a draft: see docs/protocol-format.md.

Draft it from your notes

steadyhand draft notes.txt --robot opentrons-flex -o protocol.yaml --decide "24 samples, in triplicate"

draft is the only place a language model is used. The model writes the protocol file; steadyhand then checks, plans and rehearses that file exactly as if you had written it, and sends whatever stops it back to the model, up to three times. The file lists what your text left open and the model decided under chose, and --decide settles any of it up front. More in docs/drafting.md.

What it stops, measured

A dry run that passes says little unless you know what it would have failed on.

Measured Result
Known mistakes planted in a working protocol, and stopped 24 of 26 on the Hamilton, 18 of 21 on the Flex and on the OT-2
Drafts from manufacturers' kit manuals that stand as rehearsed programs with no hand edits 121 of 146, from 71 manuals in ten sets
Automated tests 200
Runs compared with a real robot none yet

steadyhand faults runs the first row yourself; the tables are in docs/faults.md and a test pins them, so a change in either direction shows. What gets through is named there too: a tip carried down a dilution series, and a plate on the deck that is not the plate in the plan. The drafting numbers, with every manual, every caveat and what the agents found unclear, are in evals/RESULTS.md. The sets are small and were read by the project itself, not by someone who runs these kits.

Read more

The commands check, plan, rehearse, emit, faults: what each catches and decides, how to change the plan, and the replay page
The protocol file Every field: transfers, dilutions, mixes, waits, pauses, the magnet, needs, chose
Robots What differs on the Flex and the OT-2, and what Opentrons' simulator adds
Drafting from text Using your own model, what the model is told, and how a draft is fed back
Rehearsing a PyLabRobot script you already have The recording backend, and how the scene is built
What it does not do yet Read this before relying on anything here
Measurements Every drafting run, newest first

The replay page: the deck from above, a cross-section through the wells, and the command where eight tips stop on the wall of a plate

Every rehearsal writes a single replay.html that plays the run back and opens at whatever stopped it.

Development

pip install -e ".[dev]"
pytest

Earlier commits and notes call this project drybench; it was renamed at 0.14.0.

License

Apache-2.0. See LICENSE.

Metadata

Release files for steadyhand 0.37.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for steadyhand 0.37.0
File Size Uploaded Provenance Actions
steadyhand-0.37.0.tar.gz 3.0 MB With provenance Details for steadyhand-0.37.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for steadyhand 0.37.0
File Interpreter ABI Platform Actions
steadyhand-0.37.0-py3-none-any.whl Python 3 none any Details for steadyhand-0.37.0-py3-none-any.whl

Total release size: 3.1 MB

Release files / steadyhand-0.37.0.tar.gz

Download URL steadyhand-0.37.0.tar.gz
Size 3.0 MB
Tags Source
SHA-256 checksum
How to use checksums
56e4b47cfe45b0795670ca8a265e91342e3f35cbdf839831ad4d35b6119e7535
BLAKE2b-256 checksum
How to use checksums
198d004679ca4bab9c5c996d517956b5848e6e8d2aad05cc6317f8761eab28ba
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 15, 2026.

Transparency log

Release files / steadyhand-0.37.0-py3-none-any.whl

Download URL steadyhand-0.37.0-py3-none-any.whl
Size 122.6 kB
Tags Python 3
SHA-256 checksum
How to use checksums
69e84bb841223a83e30284513682072c1346d21834823675398bc664ce5bb5e8
BLAKE2b-256 checksum
How to use checksums
7066242044c392f05edd8fdbbc4e0e40b2041d9874ffce9311fcbe002787265a
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 15, 2026.

Transparency log

Security for steadyhand 0.37.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for steadyhand-0.37.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 15, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page