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

# Manage evaluation targets

The `eval target` command group lists TEA source evaluation targets and their versions. A target usually corresponds to a deployed Runtime; `eval run` resolves that object from the target ID or name and can submit an experiment against a specific target version.

<Note>
  `eval target` is supported only on the TEA evaluation backend. The default target type is `101`, the standard Volcengine source evaluation target type.
</Note>

## eval target list

List source evaluation targets, optionally filtered by a name substring.

| Flag / Argument | Description | Default |
| - | - | - |
| `--target-type <n>` | Target type. | `101` |
| `--name <name>` | Filter by a name substring. | — |
| `-r, --region <region>` | Volcengine region. | Environment |
| `--json` | Output raw JSON. | `false` |

```bash lines theme={null}
agentkit eval target list --name my-agent
```

## eval target version-list

List versions of a source evaluation target. When `eval run` omits `--target-version`, the CLI chooses the latest version from this list.

| Flag / Argument | Description | Default |
| - | - | - |
| `--source-target-id <id>` | Source evaluation target ID (required). | — |
| `--target-type <n>` | Target type. | `101` |
| `-r, --region <region>` | Volcengine region. | Environment |
| `--json` | Output raw JSON. | `false` |

```bash lines theme={null}
agentkit eval target version-list --source-target-id <source-target-id>
```

Run `agentkit eval backend` to confirm TEA and ensure the target is visible in that workspace. The source target ID for `version-list` comes from `target list`, not an experiment

```bash lines theme={null}
agentkit eval target list --name my-agent --json
agentkit eval target version-list --source-target-id <source-target-id> --json
```

Record the returned target version and pin it with `--target-version` when submitting an experiment, so later deployments do not change what “latest” means. For no matches, check the backend, workspace, and target name instead of using an unrelated resource ID
