openapi: 3.0.3
info:
  title: ''
  version: 0.0.1
paths:
  /api/v1/gpu_cluster:
    put:
      tags:
        - GPUClusterService
      summary: Update a GPU Cluster.
      operationId: GPUClusterService_Update
      requestBody:
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/GPUClusterUpdateRequest'
        required: true
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/GPUClusterUpdateResponse'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X PUT \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 --data '{ "cluster_id": "cluster id", "cluster_type": "kubernetes", "num_gpus": 24 }' \
                 https://manager.cloud.together.ai/api/v1/gpu_cluster
    post:
      tags:
        - GPUClusterService
      summary: Create GPU Cluster
      operationId: GPUClusterService_Create
      requestBody:
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/GPUClusterCreateRequest'
        required: true
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/GPUClusterCreateResponse'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X POST \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 --data '{ "region": "us-west-2", "gpu_type": "H100_SXM", "num_gpus": 8, "cluster_name": "my-gpu-cluster", "duration_days": 7, "driver_version": "CUDA_12_6_560" }' \
                 https://manager.cloud.together.ai/api/v1/gpu_cluster
  /api/v1/gpu_cluster/{cluster_id}:
    get:
      tags:
        - GPUClusterService
      summary: Get GPU cluster by cluster ID
      operationId: GPUClusterService_Get
      parameters:
        - name: cluster_id
          in: path
          required: true
          schema:
            type: string
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/GPUClusterInfo'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X GET \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 https://manager.cloud.together.ai/api/v1/gpu_cluster/${CLUSTER_ID}
    delete:
      tags:
        - GPUClusterService
      summary: Delete GPU cluster by cluster ID
      operationId: GPUClusterService_Delete
      parameters:
        - name: cluster_id
          in: path
          required: true
          schema:
            type: string
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/GPUCLusterDeleteResponse'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X DELETE \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 https://manager.cloud.together.ai/api/v1/gpu_cluster/${CLUSTER_ID}
  /api/v1/gpu_clusters:
    get:
      tags:
        - GPUClusterService
      summary: List all GPU clusters.
      operationId: GPUClusterService_List
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/GPUClusters'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X GET \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 https://manager.cloud.together.ai/api/v1/gpu_clusters
  /api/v1/regions:
    get:
      tags:
        - RegionService
      summary: List regions and corresponding supported driver versions
      operationId: RegionService_List
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/RegionListResponse'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X GET \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 https://manager.cloud.together.ai/api/v1/regions
  /api/v1/shared_volume:
    put:
      tags:
        - SharedVolumeService
      summary: Update a shared volume.
      operationId: SharedVolumeService_Update
      requestBody:
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/SharedVolumeUpdateRequest'
        required: true
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SharedVolumeInfo'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X PUT \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 --data '{ "volume_id": "12345-67890-12345-67890", "size_tib": 3}' \
                 https://manager.cloud.together.ai/api/v1/shared_volume
    post:
      tags:
        - SharedVolumeService
      summary: Create a shared volume.
      operationId: SharedVolumeService_Create
      requestBody:
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/SharedVolumeCreateRequest'
        required: true
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SharedVolumeCreateResponse'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X POST \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 --data '{ "volume_name": "my-shared-volume", "size_tib": 2, "region": "us-west-2" }' \
                 https://manager.cloud.together.ai/api/v1/shared_volume
  /api/v1/shared_volume/{volume_id}:
    get:
      tags:
        - SharedVolumeService
      summary: Get shared volume by volume Id.
      operationId: SharedVolumeService_Get
      parameters:
        - name: volume_id
          in: path
          required: true
          schema:
            type: string
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SharedVolumeInfo'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X GET \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 https://manager.cloud.together.ai/api/v1/shared_volume/${SHARED_VOLUME_ID}
    delete:
      tags:
        - SharedVolumeService
      summary: Delete shared volume by volume id.
      operationId: SharedVolumeService_Delete
      parameters:
        - name: volume_id
          in: path
          required: true
          schema:
            type: string
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SharedVolumeDeleteResponse'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X DELETE \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 https://manager.cloud.together.ai/api/v1/shared_volume/${SHARED_VOLUME_ID}
  /api/v1/shared_volumes:
    get:
      tags:
        - SharedVolumeService
      summary: List all shared volumes.
      operationId: SharedVolumeService_List
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/SharedVolumes'
      x-codeSamples:
        - lang: Shell
          label: cURL
          source: |
            curl -X GET \
                 -H "Authorization: Bearer $TOGETHER_API_KEY" \
                 https://manager.cloud.together.ai/api/v1/shared_volumes
components:
  schemas:
    GPUCLusterDeleteResponse:
      type: object
      properties:
        cluster_id:
          type: string
    GPUClusterControlPlaneNode:
      type: object
      properties:
        node_id:
          type: string
        node_name:
          type: string
        status:
          type: string
        host_name:
          type: string
        num_cpu_cores:
          type: integer
          format: uint32
        memory_gib:
          type: number
          format: float
        network:
          type: string
    GPUClusterCreateRequest:
      required:
        - region
        - gpu_type
        - num_gpus
        - cluster_name
        - duration_days
        - driver_version
        - billing_type
      type: object
      properties:
        cluster_type:
          enum:
            - UNKNOWN_TYPE
            - KUBERNETES
            - SLURM
          type: string
          description: GPU Cluster create request.
          format: enum
        region:
          type: string
          description: >-
            Region to create the GPU cluster in. Valid values are us-central-8
            and us-central-4.
        gpu_type:
          enum:
            - UNKNOWN_GPU_TYPE
            - H100_SXM
            - H200_SXM
            - RTX_6000_PCI
          type: string
          description: Type of GPU to use in the cluster
          format: enum
        num_gpus:
          type: integer
          description: >-
            Number of GPUs to allocate in the cluster. This must be multiple of
            8. For example, 8, 16 or 24
          format: int32
        cluster_name:
          type: string
          description: Name of the GPU cluster.
        duration_days:
          type: integer
          description: Duration in days to keep the cluster running.
          format: uint32
        driver_version:
          enum:
            - UNKNOWN_DRIVER
            - CUDA_12_5_555
            - CUDA_12_6_560
            - CUDA_12_6_565
            - CUDA_12_8_570
          type: string
          description: NVIDIA driver version to use in the cluster.
          format: enum
        shared_volume:
          $ref: '#/components/schemas/SharedVolumeCreateRequest'
        volume_id:
          type: string
        billing_type:
          enum:
            - UNSPECIFIED
            - RESERVED
            - ON_DEMAND
          type: string
          format: enum
      description: GPU Cluster create request
    GPUClusterCreateResponse:
      type: object
      properties:
        cluster_id:
          type: string
        status:
          enum:
            - UNKNOWN_STATUS
            - PENDING
            - QUEUED
            - WAITING_CONTROL_PLANE
            - WAITING_DATA_PLANE
            - READY
            - FAILED
          type: string
          format: enum
    GPUClusterGPUWorkerNode:
      type: object
      properties:
        node_id:
          type: string
        node_name:
          type: string
        status:
          type: string
        host_name:
          type: string
        num_cpu_cores:
          type: integer
          format: uint32
        num_gpus:
          type: integer
          format: uint32
        memory_gib:
          type: number
          format: float
        networks:
          type: array
          items:
            type: string
    GPUClusterInfo:
      type: object
      properties:
        cluster_id:
          type: string
        cluster_type:
          enum:
            - UNKNOWN_TYPE
            - KUBERNETES
            - SLURM
          type: string
          format: enum
        region:
          type: string
        gpu_type:
          enum:
            - UNKNOWN_GPU_TYPE
            - H100_SXM
            - H200_SXM
            - RTX_6000_PCI
          type: string
          format: enum
        cluster_name:
          type: string
        duration_hours:
          type: integer
          format: uint32
        driver_version:
          enum:
            - UNKNOWN_DRIVER
            - CUDA_12_5_555
            - CUDA_12_6_560
            - CUDA_12_6_565
            - CUDA_12_8_570
          type: string
          format: enum
        volumes:
          type: array
          items:
            $ref: '#/components/schemas/GPUClusterVolume'
        status:
          type: string
        control_plane_nodes:
          type: array
          items:
            $ref: '#/components/schemas/GPUClusterControlPlaneNode'
        gpu_worker_nodes:
          type: array
          items:
            $ref: '#/components/schemas/GPUClusterGPUWorkerNode'
        kube_config:
          type: string
        num_gpus:
          type: integer
          format: int32
    GPUClusterUpdateRequest:
      type: object
      properties:
        cluster_id:
          type: string
        cluster_type:
          enum:
            - UNKNOWN_TYPE
            - KUBERNETES
            - SLURM
          type: string
          format: enum
        num_gpus:
          type: integer
          format: uint32
    GPUClusterUpdateResponse:
      type: object
      properties:
        cluster_id:
          type: string
        status:
          enum:
            - UNKNOWN_STATUS
            - PENDING
            - QUEUED
            - WAITING_CONTROL_PLANE
            - WAITING_DATA_PLANE
            - READY
            - FAILED
          type: string
          format: enum
    GPUClusterVolume:
      type: object
      properties:
        volume_id:
          type: string
        volume_name:
          type: string
        size_tib:
          type: integer
          format: uint32
        status:
          type: string
    GPUClusters:
      type: object
      properties:
        clusters:
          type: array
          items:
            $ref: '#/components/schemas/GPUClusterInfo'
    Region:
      type: object
      properties:
        id:
          type: string
        name:
          type: string
        availability_zones:
          type: array
          items:
            type: string
        driver_versions:
          type: array
          items:
            type: string
    RegionListResponse:
      type: object
      properties:
        regions:
          type: array
          items:
            $ref: '#/components/schemas/Region'
    SharedVolumeCreateRequest:
      type: object
      properties:
        volume_name:
          type: string
        size_tib:
          type: integer
          description: Volume size in whole tebibytes (TiB).
          format: uint32
        region:
          type: string
    SharedVolumeCreateResponse:
      type: object
      properties:
        volume_id:
          type: string
    SharedVolumeDeleteResponse:
      type: object
      properties:
        success:
          type: boolean
    SharedVolumeInfo:
      type: object
      properties:
        volume_id:
          type: string
        volume_name:
          type: string
        size_tib:
          type: integer
          format: uint32
    SharedVolumeUpdateRequest:
      type: object
      properties:
        volume_id:
          type: string
        size_tib:
          type: integer
          format: uint32
    SharedVolumes:
      type: object
      properties:
        volumes:
          type: array
          items:
            $ref: '#/components/schemas/SharedVolumeInfo'
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
security:
  - bearerAuth: []
tags:
  - name: GPUClusterService
  - name: RegionService
  - name: SharedVolumeService