Open Source

EthoClaw — The First OpenClaw Agent for Neuroethology

Local deploy · Natural language · Open source

Open-source agent for neuroethology, pharmacology, and behavior research — from raw video to publication-ready results, fully on local compute.

v1.0.0

First release

MIT

License

Local

Compute

EthoClaw

What is EthoClaw?

EthoClaw is BayONE Life Sciences' open-source agent for neuroethology, pharmacology, and animal behavior research.

Built on OpenClaw and driven by natural language, it covers the full pipeline from raw video to publication-ready results — leveraging local compute to avoid cloud upload bottlenecks.

The preprint is on bioRxiv; full code and docs are on GitHub. Download, use, and raise issues for support.

01

Pain Points

Animal behavior analysis typically involves:

  • ·Frame-by-frame manual labeling — DeepLabCut is powerful but requires reinstalling environments on every machine
  • ·Python and MATLAB plotting is powerful, but styling often takes longer than the analysis itself
  • ·GraphPad Prism is standard, but data export, statistics, and methods writing are all manual

EthoClaw unifies these fragmented tools into one AI-driven workflow — describe what you need in plain language and get usable results.

Pain Points
02

Core Features

Zero-shot Pose Estimation
01

Zero-shot Pose Estimation

Skip frame-by-frame labeling — shoot and analyze.

Dual-mode tracking on raw video: fast CV-based detection for high-contrast scenes on CPU; high-precision DeepLabCut SuperAnimal with 27 anatomical keypoints — fully automated on a 5-minute clip.

Trajectory & Speed Heatmaps
02

Trajectory & Speed Heatmaps

Generate heatmaps via conversation — no code.

After tracking, one-click publication-grade trajectory and speed heatmaps.

Multi-group Statistics & Violin Plots
03

Multi-group Statistics & Violin Plots

Just talk to EthoClaw.

From standardized HDF5 data — automatic kinematics and group statistics: Welch's t-test for two groups, Kruskal-Wallis for three or more with post-hoc comparisons.

Radar Charts & Clustering
Radar Charts & Clustering
04

Radar Charts & Clustering

Multi-parameter group comparison

Multi-dimensional radar charts for phenotypic differences across groups, plus hierarchical clustering heatmaps.

Publication Layout
05

Publication Layout

Let EthoClaw handle layout.

Publication figure layout that used to require Illustrator — now automated by EthoClaw.

Automated Analysis Reports
06

Automated Analysis Reports

Ask EthoClaw for step-by-step details.

Structured reports with background, sample info, statistics, and conclusions — continue the conversation for computational details.

Literature Tracking & Knowledge Base
07

Literature Tracking & Knowledge Base

Stop manually checking papers every day.

Daily arXiv/PubMed digests with ranked summaries; local PDF interpretation builds a personal paper library.

03

Technical Architecture

Built on OpenClaw with conversational control and local compute — no massive cloud uploads.

  • Multi-skill stack: OpenCV fast mode + DeepLabCut SuperAnimal; standardized HDF5/CSV output
  • Optimized OpenClaw for faster response in high-frequency tool-call scenarios
  • OpenClaw ecosystem: ClawHub plugins, Feishu/Telegram remote analysis
  • NVIDIA GPU recommended; CPU-only mode available without SuperAnimal pose estimation
Technical Architecture
04

Open Source

v1.0.0 released on GitHub under MIT license — completely free.

Known limitations in the current release:

  • !Single-animal top-view only — no 3D or multi-animal ID tracking yet
  • !Tracking optimized for black mice on white backgrounds
  • !No interactive ROI drawing yet — specify coordinates via conversation
  • !AI-driven pipeline may contain errors — manual review strongly recommended

Submit GitHub Issues for feedback. Cloud demo version coming soon.

Open Source

MIT License

Get Started with EthoClaw

Download on GitHub or contact us for deployment guidance and lab integration.

GitHub