# Create an agent

Create an agent resource in the current workspace. The caller needs permission to create agents in the workspace; do not pass `workspace_id` or `user_id` in the request body, the service will determine them from the request context.

```text
POST https://moi.matrixorigin.cn/newmoi/workspaces/{workspace_id}/agents
```

## Preparation before calling

Prepare a personal access token and target workspace ID that has access to the target workspace.

The example below uses:

- `$AI_STUDIO_API_KEY`: The actual personal access token, passed through the `X-API-Key` Header.
- `$WORKSPACE_ID`: The workspace ID of the agent to be created, passed via the `X-Workspace-ID` Header and as `workspace_id` in the path.

Models, tools, skills, knowledge bases, and runtime environments referenced in the request body must be parsable.

## Request example

```bash
curl -X POST "https://moi.matrixorigin.cn/newmoi/workspaces/$WORKSPACE_ID/agents" \
  -H "X-API-Key: $AI_STUDIO_API_KEY" \
  -H "X-Workspace-ID: $WORKSPACE_ID" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "销售助手",
    "instruction": {
      "system_prompt": "分析销售数据"
    },
    "runtime": {
      "provider": "astra",
      "profile": "default"
    }
  }'
```

## Path parameters

| Parameters | Type | Is it required | Description |
| --- | --- | --- | --- |
| `workspace_id` | string | Yes | The ID of the workspace where the agent is to be created. |

## Request body

| Field | Type | Is it required | Description |
| --- | --- | --- | --- |
| `name` | string | Yes | Agent display name. |
| `instruction` | object | Yes | Command configuration, such as `system_prompt`, `behavior_rules`. |
| `runtime` | object | Yes | Run target containing `provider` and `profile`. |
| `description` | string | No | Agent description. |
| `model` | object | No | Model configuration, which can include `model_config_ref`, `default_model`, `params_override`. |
| `binding` | object | No | Resource bindings such as tools, skills, knowledge bases, and channels. |
| `policy_refs` | object | No | Operation, approval, and guardrail policy references. |
| `workflow_refs` | object | No | Workflow template reference. |
| `status` | string | No | Agent state; created as `draft` if not provided. |
| `id` | string | No | Agent ID. Generated by the server if not provided. |
| `avatar_ref` | string | No | Agent avatar and icon reference. |
| `icon` | string | No | Agent avatar and icon reference. |
| `display_tags` | string[] | No | Label used for display. |
| `category` | string | No | Agent classification. |
| `sort_order` | integer | No | Display the sort value. |
| `labels` | object | No | String key-value label and comment. |
| `annotations` | object | No | String key-value label and comment. |
| `source_type` | string | No | Agent source type: `custom` or `system`; if not provided, `custom`. |
| `source_ref` | string | No | Agent source reference. |
| `metadata` | object | No | Extended metadata. |

`[]` after a type denotes an array. `[]` in a field path denotes each item in an array.

## Successful response

Returns `201` on success. `data` is a new agent.

```json
{
  "code": 0,
  "data": {
    "id": "agent_01",
    "workspace_id": "ws_01",
    "schema_version": 1,
    "name": "销售助手",
    "instruction": {
      "system_prompt": "分析销售数据"
    },
    "runtime": {
      "provider": "matrixone",
      "profile": "default"
    },
    "status": "draft",
    "version": 1,
    "source_type": "custom",
    "created_by": "user_01",
    "updated_by": "user_01",
    "created_at": "2026-01-15T10:00:00Z",
    "updated_at": "2026-01-15T10:00:00Z"
  }
}
```

The response fields are as follows.

| Field | Type | Description |
| --- | --- | --- |
| `code` | integer | `0` on success. |
| `data.id` | string | Agent ID and workspace ID. |
| `data.workspace_id` | string | Agent ID and workspace ID. |
| `data.schema_version` | integer | Agent metadata structure version. |
| `data.name` | string | Agent name and description; will not be returned if no description is set. |
| `data.description` | string | Agent name and description; will not be returned if no description is set. |
| `data.avatar_ref` | string | Avatar reference; does not return if not set. |
| `data.icon` | string | Icon reference; does not return if not set. |
| `data.display_tags` | string[] | Display tag; does not return if not set. |
| `data.category` | string | Agent classification; does not return if not set. |
| `data.sort_order` | integer | Displays the sorting value; does not return if not set. |
| `data.instruction` | object | Command configuration. |
| `data.runtime` | object | Run target containing `provider`, `profile`, and optionally `config`. |
| `data.model` | object | Model configuration and resource binding; does not return if not set. |
| `data.binding` | object | Model configuration and resource binding; does not return if not set. |
| `data.policy_refs` | object | Policy and workflow reference; not returned if not set. |
| `data.workflow_refs` | object | Policy and workflow reference; not returned if not set. |
| `data.lifecycle` | object | Life cycle information; does not return if not set. |
| `data.status` | string | Agent status and resource version. |
| `data.version` | integer | Agent status and resource version. |
| `data.source_type` | string | Source type and source reference; not returned when source reference is not set. |
| `data.source_ref` | string | Source type and source reference; not returned when source reference is not set. |
| `data.labels` | object | Extended tags, comments, and metadata; not returned if not set. |
| `data.annotations` | object | Extended tags, comments, and metadata; not returned if not set. |
| `data.metadata` | object | Extended tags, comments, and metadata; not returned if not set. |
| `data.created_by` | string | Creator and last updater ID; not returned if not set. |
| `data.updated_by` | string | Creator and last updater ID; not returned if not set. |
| `data.created_at` | string | Creation and last update times, in RFC 3339 format. |
| `data.updated_at` | string | Creation and last update times, in RFC 3339 format. |

`[]` after a type denotes an array. `[]` in a field path denotes each item in an array.

## Error response

```json
{
  "code": 2,
  "message": "invalid agent metadata"
}
```

### Common HTTP errors

```{list-table}
:header-rows: 1
:widths: 12 22 32 34

* - HTTP status code
  - error code
  - Common causes
  - Recommended actions
* - `400`
  - `2`（`INVALID_ARGUMENT`）
  - The request body, agent field, run target, or resource reference is invalid.
  - Check request fields and referenced resources.
* - `401`
  - `6`（`UNAUTHENTICATED`）
  - Lack of valid identity credentials.
  - Check API Key.
* - `403`
  - `5`（`PERMISSION_DENIED`）
  - The current identity does not have permission to create agents in the workspace.
  - Use an authorized identity, or contact the administrator for authorization.
* - `409`
  - `4`（`ALREADY_EXISTS`）
  - The same agent ID already exists in the current workspace.
  - Use a new ID, or query an existing agent.
* - `503`
  - `15`（`UNAVAILABLE`）
  - The agent resource service or its authorized dependencies are temporarily unavailable.
  - Try again later.
```

## Follow-up operations

Log `data.id`. Use the ID to [Query agent details](get-agent.md) to confirm the saving result; when you need to view the version, [Query the agent version](list-agent-versions.md), when you need to configure binding, [Query resource binding](get-resource-bindings.md) or [Query execution strategy](get-runtime-policies.md).