Natural language AI agents for lab robot control — no coding required.
"I had 30 scientists waiting for methods while I was the only automation engineer. OpenLabAI is what I built so that never happens again." — Ainur Nygmet, ZenoVistaAI Inc.
OpenLabAI lets a wet lab scientist describe an experiment in plain English and get a ready-to-run liquid handling protocol in under 5 minutes — without writing a single line of code.
You type:
Plan an NGS library cleanup: bind 40 µL AMPure beads, 3x 80 µL ethanol wash, elute 20 µL
The agent plans it, shows you a Gantt timeline, and generates the protocol file for your robot.
In most biotech labs today, there is one automation engineer for every 20–30 scientists. Every time a scientist needs a new liquid handling method, they wait. Days. Sometimes weeks.
PyLabRobot (Wierenga et al., 2023) and Pioneer Labs (Mancuso et al., 2026) solved the programmer bottleneck — they replaced proprietary vendor software with Python. That is a huge step.
OpenLabAI solves the next bottleneck: the scientist still needs to know Python.
We add a conversational AI layer on top of PyLabRobot so that the scientist talks to the robot directly — in the language of science, not code.
| Traditional | PyLabRobot / Pioneer Labs | OpenLabAI | |
|---|---|---|---|
| Who can write protocols | Automation engineer only | Python programmers | Any scientist |
| Interface | Proprietary GUI | Jupyter notebooks | Plain English chat |
| Time to new protocol | Days–weeks | Hours | Minutes |
| Hardware support | One vendor | Hamilton, Tecan, OT-2 | + Cellario + Biomek FXP |
| AI integration | None | Code assist | Full conversational agent |
| Cost | $50k+ software | Free | Free |
| Instrument | Tier | Connection | Status |
|---|---|---|---|
| Opentrons OT-2 | Tier 1 — Full live control | HTTP API | ✅ Production |
| Hamilton STAR/STARLet | Tier 2 — COM automation | PyLabRobot USB | ✅ Production |
| Cellario workcells | Tier 2 — COM automation | COM interface | ✅ Beta |
| Beckman Biomek FXP | Tier 3 — File-based | .mth XML files | ✅ Production |
| Tecan Freedom EVO | Tier 2 — COM automation | PyLabRobot | 🔧 In progress |
Download BiomekAgent.html, open it in Chrome, paste your Claude API key, and start talking to your robot. No Python, no terminal, no installation.
Requirements: Python 3.13+, pip
# 1. Clone the repo
git clone https://github.com/nygmeta/OpenLabAI.git
cd OpenLabAI
# 2. Install dependencies
pip install -r requirements.txt
# 3. Run the MCP server for your instrument
python mcp_servers/ot2_server.py # Opentrons OT-2
python mcp_servers/biomek_server.py # Beckman Biomek FXP
python mcp_servers/cellario_server.py # Cellario workcells (Windows only)4. Add to your Claude Desktop config (%APPDATA%\Claude\claude_desktop_config.json on Windows):
{
"mcpServers": {
"openlab": {
"command": "python",
"args": ["C:/path/to/OpenLabAI/mcp_servers/ot2_server.py"]
}
}
}5. Open Claude Desktop and start talking:
Read my OT-2 deck and tell me what is loaded
Create a 50 µL transfer from well A1 to B1 across all 96 wells, new tips each row
Plan an AMPure bead cleanup: 1.8x beads, 2x 80% ethanol wash, elute in 20 µL EB
OpenLabAI/
├── gui/
│ └── BiomekAgent.html # Standalone web GUI — open in Chrome, no install
├── mcp_servers/
│ ├── ot2_server.py # Opentrons OT-2 MCP server (HTTP API)
│ ├── biomek_server.py # Beckman Biomek FXP MCP server (file-based)
│ └── cellario_server.py # Cellario workcell MCP server (Windows COM)
├── evals/
│ ├── protocol_evals.py # Acceptance criteria and validation framework
│ └── run_logger.py # Audit trail and run logging for regulated environments
├── protocols/
│ ├── ngs_cleanup.py # AMPure bead cleanup (OT-2, PyLabRobot)
│ ├── normalization.py # DNA/library normalization
│ └── serial_dilution.py # Serial dilution template
├── resources/
│ └── custom_labware.py # Custom labware definitions for PyLabRobot
├── examples/
│ ├── ngs_cleanup_example.md # Step-by-step walkthrough
│ └── slas_boston_case_study.md # SLAS Boston 2025 live demo case study
├── docs/
│ ├── INSTRUMENT_GUIDE.md # How to connect each instrument type
│ └── SCIENTIST_GUIDE.md # For scientists with no coding background
├── requirements.txt
├── LICENSE
└── README.md
You: Plan an NGS library cleanup. I have 50 µL of PCR product in slot 1.
Use 1.8x AMPure beads, 2x 80% ethanol wash, elute in 20 µL EB.
Agent: Here is your NGS cleanup protocol — 11 steps, ~28 minutes:
Step 1: Pick up tips
Step 2: Aspirate 90 µL AMPure beads from reservoir
Step 3: Dispense beads to sample plate
Step 4: Mix 10x at 80 µL to bind DNA
Step 5: Incubate 5 min on magnet
Step 6: Remove supernatant
Step 7: Ethanol wash 1 — 150 µL
Step 8: Ethanol wash 2 — 150 µL
Step 9: Air dry 2 min
Step 10: Add 20 µL EB buffer
Step 11: Transfer eluate to destination plate
You: Normalize my library plate to 4 nM. Source concentrations are in this CSV.
Agent: I will calculate transfer volumes from your concentration data
and generate a normalization worklist for the OT-2.
Scientist (plain English)
↓
Claude Agent (protocol planning + SFS tree search)
↓
MCP Server (tool calls: read_deck, aspirate, create_protocol)
↓
PyLabRobot / HTTP API / COM / File
↓
Physical Robot
OpenLabAI uses Scattered Forest Search (Light et al., 2024) to generate and validate multiple protocol candidates before committing. This increases first-attempt protocol validity from 54% (single-shot generation) to 87% (SFS).
The evals/ folder contains a production-grade validation framework for AI-generated protocols:
- Acceptance criteria per protocol type (NGS cleanup, normalization, serial dilution)
- Deck constraint checker — validates volumes, positions, tip availability per instrument
- Run logger — full audit trail with operator, timestamp, step status, and protocol hash
- Designed for GxP-adjacent environments requiring traceability
from evals.protocol_evals import evaluate_protocol
result = evaluate_protocol(protocol, protocol_type="ngs_cleanup", instrument="OT-2")
print(result.overall_score) # 0.87
print(result.passed) # True
print(result.protocol_hash) # abc123...If you use OpenLabAI in your research, please cite:
@article{nygmet2026openlabai,
title={Natural Language Agents for Laboratory Automation: An MCP-Based Framework
for Scientist-Directed Robot Control Without Coding},
author={Nygmet, Ainur},
journal={bioRxiv},
year={2026},
institution={ZenoVistaAI Inc.}
}This work builds on:
- PyLabRobot — Wierenga et al., 2023
- Scattered Forest Search — Light et al., 2024
- Pioneer Labs NGS Library Prep — Mancuso et al., 2026
We welcome contributions — especially from wet lab scientists who can tell us what is missing.
- Found a bug? Open a GitHub Issue
- New instrument backend? Open a PR with your MCP server
- New labware definitions? Add to
resources/custom_labware.py - Protocol templates? Add to
protocols/
Built by Ainur Nygmet at ZenoVistaAI Inc.
Background: 6+ years as a lab automation engineer at Guardant Health, Personalis, and Hexagon Bio. Developed 50+ Hamilton methods, trained 40+ scientists on lab automation, certified Cellario operator. Demonstrated live AI-driven workcell orchestration at SLAS Boston 2025.
This project exists because I was the bottleneck. I do not want anyone else to be.
Contact: nygmetainur@gmail.com LinkedIn: Ainur Nygmet GitHub: nygmeta
MIT License — free to use, modify, and share. See LICENSE for details.