TensorWave Open AI API

openapi: 3.2.0
info:
  title: ScalarLM Open AI API
  version: '1.151'
  summary: Unified LLM inference and training API from the ScalarLM stack, maintained and sponsored by TensorWave.
  description: 'ScalarLM is TensorWave''s open-source (CC0-1.0) unified training and inference stack. A single deployment exposes an OpenAI-compatible inference endpoint backed by vLLM, a queue-backed batch generate surface, a Megatron-LM training surface dispatched through Slurm, and health/observability endpoints.


    PROVENANCE: this document is DERIVED by API Evangelist from the first-party FastAPI source in github.com/tensorwavecloud/ScalarLM. Every path, method, parameter and schema below was read from the committed route decorators and Pydantic models listed in info.x-evidence.sources. TensorWave does not publish a static OpenAPI document; a live ScalarLM deployment serves FastAPI''s own generated spec at /openapi.json. This is not a provider-published artifact, and the operationIds here are API Evangelist naming rather than FastAPI''s generated ids.'
  license:
    name: CC0-1.0
    url: https://creativecommons.org/publicdomain/zero/1.0/
  contact:
    name: TensorWave
    url: https://tensorwave.com/connect
  x-evidence:
    method: derived
    derived_from: source-code
    derived_on: '2026-08-02'
    repository: https://github.com/tensorwavecloud/ScalarLM
    repository_license: CC0-1.0
    sources:
    - https://raw.githubusercontent.com/tensorwavecloud/ScalarLM/main/infra/cray_infra/api/fastapi/main.py
    - https://raw.githubusercontent.com/tensorwavecloud/ScalarLM/main/infra/cray_infra/api/fastapi/routers/openai_v1_router.py
    - https://raw.githubusercontent.com/tensorwavecloud/ScalarLM/main/infra/cray_infra/api/fastapi/routers/megatron_router.py
    - https://raw.githubusercontent.com/tensorwavecloud/ScalarLM/main/infra/cray_infra/api/fastapi/routers/health_router.py
    - https://raw.githubusercontent.com/tensorwavecloud/ScalarLM/main/infra/cray_infra/api/fastapi/routers/generate_router.py
    - https://raw.githubusercontent.com/tensorwavecloud/ScalarLM/main/infra/cray_infra/api/fastapi/routers/slurm_router.py
    - https://raw.githubusercontent.com/tensorwavecloud/ScalarLM/main/infra/cray_infra/api/fastapi/routers/request_types/
    note: Route table read from the FastAPI APIRouter prefixes in main.py (openai_v1_router, megatron_router '/megatron', health_router '/health' and generate_router '/generate' are all mounted under '/v1'; slurm_router '/slurm' is mounted at the app root). Schemas transcribed from the Pydantic models under routers/request_types/. CompletionRequest and ChatCompletionRequest are vLLM types re-exported by ScalarLM and are modelled as open objects rather than reproduced field by field.
    live_spec_hint: <deployment>/openapi.json (FastAPI generated)
    documented_deployment:
      url: https://gpt-oss.cray-lm.com
      source: https://www.scalarlm.com/inference/
      probed: '2026-08-02'
      http_status: 530
      note: Documented demo deployment; origin unreachable (Cloudflare 530) at probe time.
servers:
- url: https://gpt-oss.cray-lm.com
  description: Public ScalarLM demo deployment documented at scalarlm.com/inference (origin returned HTTP 530 when probed 2026-08-02).
- url: http://localhost:8000
  description: Local development server started by ./scalarlm up.
tags:
- name: OpenAI
  description: OpenAI-compatible inference endpoints proxied to vLLM.
paths:
  /v1/models:
    get:
      operationId: listModels
      summary: List available models
      tags:
      - OpenAI
      description: Proxies the upstream vLLM server's /v1/models. Returns the OpenAI model list envelope.
      responses:
        '200':
          description: Successful response
          content:
            application/json:
              schema:
                type: object
                additionalProperties: true
        '422':
          description: Validation error (FastAPI request validation).
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /v1/completions:
    post:
      operationId: createCompletion
      summary: Create a completion
      tags:
      - OpenAI
      description: OpenAI-compatible text completion. Proxied to the vLLM server; the request body is a vLLM CompletionRequest (OpenAI completions schema). Responses stream as text/event-stream.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/CompletionRequest'
      responses:
        '200':
          description: Server-sent event stream of completion chunks.
          content:
            text/event-stream:
              schema:
                type: string
        '422':
          description: Validation error (FastAPI request validation).
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
  /v1/chat/completions:
    post:
      operationId: createChatCompletion
      summary: Create a chat completion
      tags:
      - OpenAI
      description: OpenAI-compatible chat completion. When stream is true the request is proxied straight to vLLM as an SSE stream; when false it is admitted to the queue-backed path (admission control -> coalescer -> SQLite inference work queue -> worker -> result router) and answered with a whitespace-heartbeat chunked JSON response.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ChatCompletionRequest'
      responses:
        '200':
          description: Chat completion, streamed (SSE) or chunked JSON.
          content:
            application/json:
              schema:
                type: object
                additionalProperties: true
        '422':
          description: Validation error (FastAPI request validation).
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/HTTPValidationError'
components:
  schemas:
    ChatCompletionRequest:
      type: object
      required:
      - messages
      description: vLLM's OpenAI-compatible ChatCompletionRequest, imported by the router from vllm.entrypoints.openai.chat_completion.protocol. The authoritative field list is vLLM's, not ScalarLM's. stream selects the direct SSE proxy path vs the queue-backed path.
      properties:
        model:
          type: string
        messages:
          type: array
          items:
            $ref: '#/components/schemas/ChatMessage'
        max_tokens:
          type: integer
        temperature:
          type: number
        stream:
          type: boolean
        tools:
          type: array
          items:
            type: object
      additionalProperties: true
    ValidationError:
      type: object
      required:
      - loc
      - msg
      - type
      properties:
        loc:
          type: array
          items:
            oneOf:
            - type: string
            - type: integer
        msg:
          type: string
        type:
          type: string
    HTTPValidationError:
      type: object
      properties:
        detail:
          type: array
          items:
            $ref: '#/components/schemas/ValidationError'
    ChatMessage:
      type: object
      required:
      - role
      - content
      properties:
        role:
          type: string
          enum:
          - system
          - user
          - assistant
          - tool
        content:
          type: string
    CompletionRequest:
      type: object
      required:
      - prompt
      description: vLLM's OpenAI-compatible CompletionRequest, imported by the router from vllm.entrypoints.openai.completion.protocol. The authoritative field list is vLLM's, not ScalarLM's; only the fields on the router's allow-list are forwarded upstream.
      properties:
        model:
          type: string
        prompt:
          oneOf:
          - type: string
          - type: array
            items:
              type: string
        max_tokens:
          type: integer
        temperature:
          type: number
        stream:
          type: boolean
      additionalProperties: true