> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.conductorquantum.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.conductorquantum.com/_mcp/server.

# Agents

> AI-powered calibration agents built on NVIDIA Ising Calibration for autonomous quantum processor tune-up.

## NVIDIA Ising Calibration

[NVIDIA Ising Calibration 1](https://huggingface.co/nvidia/Ising-Calibration-1-35B-A3B) is part of the [NVIDIA Ising](https://www.nvidia.com/en-us/solutions/quantum-computing/ising/) open model family, the first AI models purpose-built for quantum computing workloads. Conductor Quantum contributed to the model and the accompanying benchmark.

Ising Calibration 1 is a 35B-parameter Mixture-of-Experts vision-language model (MoE VLM) fine-tuned to interpret quantum calibration experiment plots and infer the actions needed to tune a quantum processor. It analyses plot images from characterisation experiments (spectroscopy sweeps, Rabi oscillations, readout diagnostics, and more) and produces structured technical assessments: what the experiment shows, whether it succeeded, what parameters to extract, and what to do next.

The model connects directly to an agent to fully automate QPU calibration loops, replacing hours of manual plot inspection with reliable, scalable AI analysis. It is available through the [Control API](/api-reference/control-api/agents) and open weights are on [Hugging Face](https://huggingface.co/nvidia/Ising-Calibration-1-35B-A3B).

## Usage

### List available agents

**`Python`**

```python title="Python"
from conductorquantum import ConductorQuantum

client = ConductorQuantum(token="YOUR_TOKEN")

agents = client.control.agents.list()
for agent in agents:
    print(f"{agent.id}: {agent.name}")
```

### Run Ising Calibration on a plot image

Send a base64-encoded calibration plot image and an optional prompt. The agent returns a structured assessment of the experiment.

**`Python`**

```python title="Python"
import base64
from conductorquantum import ConductorQuantum

client = ConductorQuantum(token="YOUR_TOKEN")

with open("calibration_plot.png", "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode()

result = client.control.agents.run(
    "ising-calibration-v1",
    body={
        "image_base64": image_b64,
        "prompt": "Analyse this Rabi oscillation plot and extract the pi-pulse duration.",
        "max_tokens": 256,
    },
)

print(result)
```

### Async usage

**`Python`**

```python title="Python"
import asyncio
import base64
from conductorquantum import AsyncConductorQuantum

client = AsyncConductorQuantum(token="YOUR_TOKEN")


async def main():
    with open("calibration_plot.png", "rb") as f:
        image_b64 = base64.b64encode(f.read()).decode()

    result = await client.control.agents.run(
        "ising-calibration-v1",
        body={"image_base64": image_b64},
    )
    print(result)


asyncio.run(main())
```

### Request body

| Field          | Type      | Required | Description                                                                        |
| -------------- | --------- | -------- | ---------------------------------------------------------------------------------- |
| `image_base64` | `string`  | Yes      | Base64-encoded PNG or JPEG image of the calibration plot. Must be RGB (3-channel). |
| `prompt`       | `string`  | No       | Custom prompt to guide the model's analysis.                                       |
| `max_tokens`   | `integer` | No       | Maximum tokens in the response (default: model-dependent).                         |

> **Info**
>
> The VLM endpoint may experience cold-start latency of several minutes on first invocation. Subsequent requests are typically fast.

## QCalEval benchmark

The [QCalEval benchmark paper](https://research.nvidia.com/publication/2026-04_qcaleval-benchmarking-vision-language-models-quantum-calibration-plot) evaluates how well VLMs understand quantum calibration plots. Conductor Quantum contributed alongside NVIDIA and collaborators across national laboratories, universities, and quantum hardware companies. QCalEval is the first comprehensive benchmark for this domain: 243 samples across 87 scenario types from 22 experiment families, spanning superconducting qubits and neutral atoms, evaluated on six question types.

Ising Calibration 1 achieves a **74.7 zero-shot average score** on QCalEval, outperforming every general-purpose model tested, including GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6. The largest gains come on the tasks that matter most for autonomous calibration: experimental conclusion (+27.2 over base), fit quality assessment (+37.8), and calibration diagnosis (+24.7).

![QCalEval leaderboard showing top 5 models by zero-shot average score](/_fern-img/71c12914ed9445ef9599eea22b2229007e69e06af42521b5316ba1eade34d836.webp)