> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bunkerbio.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Run Inference on a Bunker Longevity Model

> Send prediction requests to a deployed Bunker Longevity model instance. Covers the predict endpoint, request shape, streaming responses, and input schemas.

Once a Bunker Longevity model is deployed and reports `ready`, you send prediction requests to the predict endpoint. The request shape is uniform across models; the payload inside each request follows the input schema documented on each model's page.

## Predict

```http theme={null}
POST /v1/models/{model}/predict
```

<CodeGroup>
  ```bash cURL theme={null}
  curl -X POST https://api.bunkerbio.com/v1/models/bunkerfold-3/predict \
    -H "Authorization: Bearer $BUNKER_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{"sequence": "MVLSPADKTNVKAAWGKVGAHAGEYGAEALER"}'
  ```

  ```python Python theme={null}
  import requests, os

  r = requests.post(
      "https://api.bunkerbio.com/v1/models/bunkerfold-3/predict",
      headers={"Authorization": f"Bearer {os.environ['BUNKER_API_KEY']}"},
      json={"sequence": "MVLSPADKTNVKAAWGKVGAHAGEYGAEALER"},
  )
  print(r.json())
  ```
</CodeGroup>

## Request shape

The request body is model-specific. See the per-model pages for input fields:

* [BunkerFold-3](/models/bunkerfold-3)
* [GenoScan-LLM](/models/genoscan-llm)
* [MolGen-7](/models/molgen-7)
* [CryoNet](/models/cryonet)
* [ToxPredict-v2](/models/toxpredict-v2)
* [EpiMark](/models/epimark)

All models accept an optional `request_id` string that is echoed back in the response for tracing.

## Response shape

Every response includes at minimum a top-level result object plus timing metadata:

```json theme={null}
{
  "request_id": "req_9e12ab",
  "model": "bunkerfold-3",
  "time_seconds": 1.8,
  "result": {
    "confidence": 0.97,
    "structure_url": "https://cdn.bunkerbio.com/predictions/abc123.pdb"
  }
}
```

<ResponseField name="request_id" type="string">
  Echo of the client-supplied `request_id`, or a server-assigned one when not provided.
</ResponseField>

<ResponseField name="time_seconds" type="number">
  Wall-clock time to produce the prediction, from receipt to response.
</ResponseField>

<ResponseField name="result" type="object">
  Model-specific result. See each model's page for its response fields.
</ResponseField>

## Async predictions

For long-running jobs (for example, [CryoNet](/models/cryonet) reconstructions), pass `mode: "async"` to receive a job ID immediately and poll for completion:

```json Request theme={null}
{
  "mode": "async",
  "particles": 12400,
  "input_url": "s3://my-bucket/particles.star"
}
```

```json Response theme={null}
{
  "job_id": "job_5c81f0",
  "status": "queued",
  "poll_url": "https://api.bunkerbio.com/v1/jobs/job_5c81f0"
}
```

Poll `poll_url` until `status` is `completed`, then read the `result` field.

## Batching

Send an array of items in a `batch` field to run multiple predictions in a single request. Batches share GPU memory and are typically faster than sequential calls.

```json theme={null}
{
  "batch": [
    {"sequence": "MVLSPADK..."},
    {"sequence": "GVEQMTGIN..."}
  ]
}
```

The response contains a `batch` array in the same order as the request.