Agentic Quantum Research

TensorCircuit-NG is an AI-native quantum research runtime for agentic discovery, algorithm design, quantum simulation, execution optimization, validation, and research artifact generation.

From published quantum knowledge to executable research infrastructure, TensorCircuit-NG connects AI agents to the context, rules, skills, and high-performance execution backends required for serious quantum research.

🚀 Experience Agent-Native Discovery

Explore how AI agents can discover, implement, optimize, validate, and communicate quantum research workflows in TensorCircuit-NG.

agent_landing/index.html

Platform Overview

Review the broader TensorCircuit-NG architecture, performance evidence, and adoption landscape.

platform/index.html

Agentic evaluation

The ORBIT-Q benchmark is the evaluation layer for agentic quantum software: it tests autonomous agents on framework-constrained scientific programming, correctness, and artifact efficiency. It makes agent-friendliness measurable rather than treating it as a code-generation slogan, while also evaluating the runtime performance of the resulting scientific artifacts.

In the reported framework-axis comparison, TensorCircuit-NG achieved the highest agent solve success rate and the lowest artifact-runtime slowdown among the evaluated quantum software frameworks. Together, the agentic skills and ORBIT-Q evaluation establish a measurable loop for building, testing, and improving AI-agent-compatible quantum software. See the accompanying paper for the benchmark design and methodology.

Why Work Within the Repository?

To write scripts and applications efficiently with AI coding agents (e.g., ClaudeCode, Cursor, Codex, Antigravity, OpenCode), we strongly recommend working directly within the local repository.

1. Rich Context: The repository contains over 100 scripts in examples/ and extensive test cases in tests/. These provide essential references that significantly reduce AI hallucinations and help the agent understand idiomatic usage.

2. Built-in Rules: We provide a dedicated AGENTS.md file. It serves as the “handbook” (similar to CLAUDE.md) for AI agents, defining coding standards and best practices.

3. Specialized Agentic Skills: The .agents/skills/ directory contains specialized workflows to guide AI assistants on complex, multi-step tasks.

Specialized Agentic Skills

TensorCircuit-NG includes built-in agentic skills that can be activated by compatible AI agents to perform advanced tasks:

• arxiv-reproduce: Translates published methods into standardized, runnable, and validated research artifacts.

• performance-optimize: Scientific execution and memory optimization workflow (JAX scanning, vectorized parallelism, etc.).

• tc-rosetta: End-to-end framework translation (from Qiskit, PennyLane, etc.) with intrinsic mathematical intent rewriting.

• tutorial-crafter: Transforms raw scripts into comprehensive, narrative-driven educational tutorials.

• demo-generator: Transforms scripts into interactive, high-performance Streamlit GUI applications.

• code-reviewer: Autonomously reviews and refactors code for mathematical correctness and performance.

• sanity-checker: Systematic audit and refactoring to reduce technical debt, improve abstractions, and ensure codebase health.

• meta-explorer: High-intensity autonomous research agent for circuit architecture and optimization strategy discovery (VQE, QML, QAOA, etc.).

Recommended Workflow

1. Clone the repository:

   git clone https://github.com/tensorcircuit/tensorcircuit-ng.git
   

2. Switch to a local playground branch:

   git checkout -b my-playground
   

3. Open the repository folder in your AI IDE: Start writing TC-NG-based scripts using natural language instructions.

By integrating extreme performance with an autonomous, intent-driven AI workflow, TensorCircuit-NG helps researchers move from a research question to implementation, exploration, optimization, validation, and a reusable scientific artifact.

Initial Prompt for Empty Workspace Setup

When starting in a completely empty folder/workspace with an AI agent, you can copy and paste the following initial prompt to bootstrap the agent. This instructs the agent to clone the repository, read coding guidelines/skills, and execute tasks accordingly:

You are an AI coding assistant. Please follow these steps to set up our project workspace and solve my requests:

1. Run `git clone https://github.com/tensorcircuit/tensorcircuit-ng.git .` to download the repository in this empty folder.
2. Install the git-cloned TensorCircuit-NG library locally in the appropriate Python environment. If unsure which Python environment is appropriate, ask the user.
3. Read `AGENTS.md` at the root of the repository to understand the coding standards.
4. Review the available agentic workflow scripts and skill files under `.agents/skills/` to see if a specialized workflow is suitable for the current task.
5. Reference existing implementations in `examples/` and test cases in `tests/` to write differentiable, and JIT-friendly code.
6. Respond to my later requests, write code, and execute experiments/tests to analyse the results.

Let me know when the initialization is complete and ask for my requests.

AI-Native Documentation

One can also refer to AI-native docs for tensorcircuit-ng: Devin Deepwiki, Google Code Wiki, and Context7 MCP.