TOOLKIT
Neutral-Atom Agentic Toolkit
Agent skills that turn a paper, a patent or your own idea into a validated experiment on a neutral-atom quantum computer.
You keep every scientific decision.
Closing the gap between ideas and quantum hardware
The Neutral-Atom Agentic Toolkit is a Pasqal add-on for AI coding agents such as Claude Code, Codex and Cursor. It takes researchers from a scientific idea to a neutral-atom experiment without first mastering neutral-atom quantum programming.
Start with a paper, a patent or your own idea. The toolkit writes the experiment spec, generates self-testing code to implement it on a neutral-atom quantum processor, validates it in device-aware emulation and, once you approve the shot count, submits it to a Pasqal QPU and analyses the results. You keep the science: the objective, the assumptions, the reading of the results and the decision to spend.
Five steps, a file you can read at each one
Describe what you want to measure, or hand over a paper. Each step leaves a file you can review.
1
Specify
- Turn your research input into a structured experiment specification. Open questions are identified for your review.
2
Implement
- Generate a self-testing Pulser sequence from the specification you approved.
3
Emulate
- Test the expected behaviour under noiseless and noisy conditions. Check the workflow implementation before using QPU time.
4
Soumettre
- Prepare and submit the experiment to a Pasqal QPU through the cloud, only after you approve the shot count.
5
Analyse
- Retrieve the measurements and compare your chosen observable with the emulated prediction.
Users can review and intervene at every step. They can edit any output and rerun the relevant stages to keep the experiment aligned with their scientific intent.
The agent orchestrates. The science stays yours.
The agent handles
- Device limits, register and pulse code
- Emulation and submission
- Collecting and comparing results
You keep
- The objective and the assumptions
- Whether the observable resolves the physics
- The reading of results and the decision to spend
No shot spent without your approval. Emulation comes before any hardware recommendation, and nothing is submitted until you approve the shot count.
Pull quote: A workflow that runs cleanly is not evidence that the physics is right. The toolkit is built around that.
Read the full technical story
Réduire les obstacles à la mise en œuvre de l'informatique quantique à atomes neutres grâce à des flux de travail autonomes
arXiv:2607.25834
Authors: C. Dalyac, A. Dauphin, L. Henriet, C. Jurczak (2026)
Turning ideas into QPU experiments: how agentic workflows make quantum computing more accessible
The three test cases in detail, where expert review was needed, and what 526 papers say about hardware reach.
Neutral-Atom Agentic Toolkit documentation
Install, configure and run the toolkit, with every command tested.
In the news
FAQ
- Closing the gap between ideas and quantum hardware
- Five steps, a file you can read at each one
- The agent orchestrates. The science stays yours.
- Read the full technical story
- In the news
- FAQ
- Neutral-Atom Agentic Toolkit
- Who is it for?
- Do I need to know Pulser?
- Which coding agents does it work with?
- What do I need to get started?
- Can QoolQit programs be emulated?
- Can the agent submit to a QPU without my approval?
- Which QPUs can I run on?
- What does Pasqal Cloud add to the toolkit?
- Can I run it through an HPC center?
- Has it been tested on real hardware?
- Can I re-analyse or publish results without Pasqal software?
Neutral-Atom Agentic Toolkit
Who is it for?
Researchers who can formulate and judge a scientific question and are comfortable with Python and a coding agent. Whether you run a protocol of your own or reproduce published work, you bring the physics and the agent handles the hardware-facing work.
Do I need to know Pulser?
No. The agent writes the Pulser code and you review it. You should be comfortable with Python and a coding agent.
Which coding agents does it work with?
Claude Code, Codex, Cursor, VS Code with Copilot, Gemini CLI, Kimi Code, DeepSeek Harness and OpenCode. Any other agent can use it via git clone and the AGENTS.md file.
What do I need to get started?
Python and Pulser are enough to specify, generate and emulate an experiment locally, with no account. Emulating at full register size needs a Pasqal Cloud account, and running on a QPU needs QPU access.
Can QoolQit programs be emulated?
Yes. QoolQit programs can be developed and validated with emulators before execution on quantum hardware. The Pasqal Software Stack supports local emulation for development on a laptop, as well as remote emulation resources through Pasqal Cloud for larger-scale simulation needs
Can the agent submit to a QPU without my approval?
No. The submission script prints the full shot count and stops until you approve it, and emulation must come first. Both rules are enforced in the scripts, not left to the model.
Which QPUs can I run on?
Pasqal QPUs available through Pasqal Cloud.
What does Pasqal Cloud add to the toolkit?
Pasqal Cloud takes you beyond local emulation: the cloud emulator at full register size, noise emulation with the live device noise model, and calibrated submission to Pasqal QPUs such as FC1 and SA1, using calibration data from the device’s current operating point. The toolkit uses the Pasqal Cloud SDK for its cloud-facing steps, through the same standard accounts external users already have.
Can I run it through an HPC center?
Yes. The submit-via-hpc route runs parametric experiments on a QPU behind an HPC cluster over SSH, with an onboarding guide for the reference site, TGCC.
Has it been tested on real hardware?
Yes. The method was run end to end on FC1 (Sherbrooke, Québec) and SA1 (Saudi Arabia), through standard Pasqal Cloud accounts. It was tested on three starting points: a published experiment reproduced on hardware, a theory paper turned into a QPU experiment, and a patent run as a multi-round procedure. A scan of the literature found that 258 of 526 neutral-atom theory papers analyzed are assessed as implementable on publicly available Pasqal QPUs today, directly or with modest adaptation. Source: arXiv:2607.25834 (2026).
Can I re-analyse or publish results without Pasqal software?
Yes. Every step leaves plain files on your disk, and analysis takes shot data only, so re-analysing, replotting or publishing a run never needs a vendor library.