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

# cyberwave.yml Manifest

> Schema reference for cyberwave.yml configuration files

## Overview

Every Cyberwave Cloud Node and edge deployment is configured via a `cyberwave.yml` manifest file. The manifest declares install scripts, inference/training commands, worker modules, hardware requirements, and MQTT settings.

## Minimal example

```yaml theme={null}
cyberwave:
  inference: inference.py
  gpu: true
```

## Full field reference

| Field          | Type           | Default     | Description                                   |
| -------------- | -------------- | ----------- | --------------------------------------------- |
| `version`      | string         | `"1"`       | Schema version                                |
| `name`         | string         | —           | Node display name                             |
| `install`      | string         | —           | Shell command run once at startup             |
| `inference`    | string         | —           | Module (`.py`) or shell command for inference |
| `training`     | string         | —           | Module (`.py`) or shell command for training  |
| `workers`      | list           | —           | Worker `.py` files using hook decorators      |
| `requirements` | list           | —           | pip package specs                             |
| `models`       | list           | —           | Model IDs to pre-download                     |
| `input`        | string or list | —           | Input type declaration                        |
| `gpu`          | boolean        | `false`     | Whether GPU is required                       |
| `runtime`      | string         | —           | Known runtime ID for zero-code mode           |
| `model`        | string         | —           | Model file path (used with `runtime`)         |
| `profile_slug` | string         | `"default"` | Profile for workload routing                  |
| `resources`    | object         | —           | `memory` and `cpus` constraints               |

## Dispatch modes

The `inference` and `training` fields auto-detect their dispatch mode:

* **Module mode** (`inference.py`): imports the file and calls `infer(**params)` directly
* **Shell mode** (`python server.py {body}`): spawns a subprocess (existing behaviour)

## Validating locally

```bash theme={null}
cyberwave manifest validate
cyberwave manifest validate path/to/cyberwave.yml --lenient
```

## Legacy compatibility

Both `cyberwave:` and `cyberwave-cloud-node:` wrapper keys are supported. The `install_script:` field (legacy) is normalised alongside `install:` (new).