llama-stack-mirror/docs/static/stainless-config.yaml

# yaml-language-server: $schema=https://app.stainlessapi.com/config-internal.schema.json

organization:
  # Name of your organization or company, used to determine the name of the client
  # and headings.
  name: llama-stack-client
  docs: https://llama-stack.readthedocs.io/en/latest/
  contact: llamastack@meta.com
security:
  - {}
  - BearerAuth: []
security_schemes:
  BearerAuth:
    type: http
    scheme: bearer
# `targets` define the output targets and their customization options, such as
# whether to emit the Node SDK and what it's package name should be.
targets:
  node:
    package_name: llama-stack-client
    production_repo: llamastack/llama-stack-client-typescript
    publish:
      npm: false
  python:
    package_name: llama_stack_client
    production_repo: llamastack/llama-stack-client-python
    options:
      use_uv: true
    publish:
      pypi: true
    project_name: llama_stack_client
  kotlin:
    reverse_domain: com.llama_stack_client.api
    production_repo: null
    publish:
      maven: false
  go:
    package_name: llama-stack-client
    production_repo: llamastack/llama-stack-client-go
    options:
      enable_v2: true
      back_compat_use_shared_package: false

# `client_settings` define settings for the API client, such as extra constructor
# arguments (used for authentication), retry behavior, idempotency, etc.
client_settings:
  default_env_prefix: LLAMA_STACK_CLIENT
  opts:
    api_key:
      type: string
      read_env: LLAMA_STACK_CLIENT_API_KEY
      auth: { security_scheme: BearerAuth }
      nullable: true

# `environments` are a map of the name of the environment (e.g. "sandbox",
# "production") to the corresponding url to use.
environments:
  production: http://any-hosted-llama-stack.com

# `pagination` defines [pagination schemes] which provides a template to match
# endpoints and generate next-page and auto-pagination helpers in the SDKs.
pagination:
  - name: datasets_iterrows
    type: offset
    request:
      dataset_id:
        type: string
      start_index:
        type: integer
        x-stainless-pagination-property:
          purpose: offset_count_param
      limit:
        type: integer
    response:
      data:
        type: array
        items:
          type: object
      next_index:
        type: integer
        x-stainless-pagination-property:
          purpose: offset_count_start_field
  - name: openai_cursor_page
    type: cursor
    request:
      limit:
        type: integer
      after:
        type: string
        x-stainless-pagination-property:
          purpose: next_cursor_param
    response:
      data:
        type: array
        items: {}
      has_more:
        type: boolean
      last_id:
        type: string
        x-stainless-pagination-property:
          purpose: next_cursor_field
# `resources` define the structure and organziation for your API, such as how
# methods and models are grouped together and accessed. See the [configuration
# guide] for more information.
#
# [configuration guide]:
#   https://app.stainlessapi.com/docs/guides/configure#resources
resources:
  $shared:
    models:
      agent_config: AgentConfig
      interleaved_content_item: InterleavedContentItem
      interleaved_content: InterleavedContent
      batch_completion: BatchCompletionResponse
      param_type: ParamType
      safety_violation: SafetyViolation
      sampling_params: SamplingParams
      scoring_result: ScoringResult
      message: Message
      user_message: UserMessage
      completion_message: CompletionMessage
      tool_response_message: ToolResponseMessage
      system_message: SystemMessage
      tool_call: ToolCall
      metric: MetricInResponse
      tool_param_definition: ToolParamDefinition
      content_delta: ContentDelta
      query_result: RAGQueryResult
      document: RAGDocument
      query_config: RAGQueryConfig
      query_generator_config: RAGQueryGeneratorConfig
      response_format: ResponseFormat
      chat_completion_response: ChatCompletionResponse
  toolgroups:
    models:
      tool_group: ToolGroup
      list_tool_groups_response: ListToolGroupsResponse
    methods:
      register: post /v1/toolgroups
      get: get /v1/toolgroups/{toolgroup_id}
      list: get /v1/toolgroups
      unregister: delete /v1/toolgroups/{toolgroup_id}
  tools:
    models:
      tool: Tool
      list_tools_response: ListToolsResponse
    methods:
      get: get /v1/tools/{tool_name}
      list:
        endpoint: get /v1/tools
        paginated: false

  tool_runtime:
    models:
      tool_def: ToolDef
      tool_invocation_result: ToolInvocationResult
    methods:
      list_tools:
        endpoint: get /v1/tool-runtime/list-tools
        paginated: false
      invoke_tool: post /v1/tool-runtime/invoke
    subresources:
      rag_tool:
        methods:
          insert: post /v1/tool-runtime/rag-tool/insert
          query: post /v1/tool-runtime/rag-tool/query

  responses:
    models:
      response_object_stream: OpenAIResponseObjectStream
      response_object: OpenAIResponseObject
    methods:
      create:
        type: http
        endpoint: post /v1/openai/v1/responses
        streaming:
          stream_event_model: responses.response_object_stream
          param_discriminator: stream
      retrieve: get /v1/openai/v1/responses/{response_id}
      list:
        type: http
        endpoint: get /v1/openai/v1/responses
      delete:
        type: http
        endpoint: delete /v1/openai/v1/responses/{response_id}
    subresources:
      input_items:
        methods:
          list:
            type: http
            endpoint: get /v1/openai/v1/responses/{response_id}/input_items

  agents:
    # Configure the methods defined in this resource. Each key in the object is the
    # name of the method and the value is either an endpoint (for example, `get /v1/foo`)
    # or an object with more detail.
    #
    # [reference]: https://app.stainlessapi.com/docs/reference/config#method
    methods:
      create: post /v1/agents
      list: get /v1/agents
      retrieve: get /v1/agents/{agent_id}
      delete: delete /v1/agents/{agent_id}
    models:
      inference_step: InferenceStep
      tool_execution_step: ToolExecutionStep
      tool_response: ToolResponse
      shield_call_step: ShieldCallStep
      memory_retrieval_step: MemoryRetrievalStep
    # Subresources define resources that are nested within another for more powerful
    # logical groupings, e.g. `cards.payments`.
    subresources:
      session:
        # Configure the models--named types--defined in the resource. Each key in the
        # object is the name of the model and the value is either the name of a schema in
        # `#/components/schemas` or an object with more detail.
        #
        # [reference]: https://app.stainlessapi.com/docs/reference/config#model
        models:
          session: Session
        methods:
          list: get /v1/agents/{agent_id}/sessions
          create: post /v1/agents/{agent_id}/session
          delete: delete /v1/agents/{agent_id}/session/{session_id}
          retrieve: get /v1/agents/{agent_id}/session/{session_id}
      steps:
        methods:
          retrieve: get /v1/agents/{agent_id}/session/{session_id}/turn/{turn_id}/step/{step_id}
      turn:
        models:
          turn: Turn
          turn_response_event: AgentTurnResponseEvent
          turn_response_event_payload: AgentTurnResponseEventPayload
          agent_turn_response_stream_chunk: AgentTurnResponseStreamChunk
        methods:
          create:
            type: http
            endpoint: post /v1/agents/{agent_id}/session/{session_id}/turn
            streaming:
              stream_event_model: agents.turn.agent_turn_response_stream_chunk
              param_discriminator: stream
          retrieve: get /v1/agents/{agent_id}/session/{session_id}/turn/{turn_id}
          resume:
            type: http
            endpoint: post /v1/agents/{agent_id}/session/{session_id}/turn/{turn_id}/resume
            streaming:
              stream_event_model: agents.turn.agent_turn_response_stream_chunk
              param_discriminator: stream

  datasets:
    models:
      list_datasets_response: ListDatasetsResponse
    methods:
      register: post /v1/datasets
      retrieve: get /v1/datasets/{dataset_id}
      list:
        endpoint: get /v1/datasets
        paginated: false
      unregister: delete /v1/datasets/{dataset_id}
      iterrows: get /v1/datasetio/iterrows/{dataset_id}
      appendrows: post /v1/datasetio/append-rows/{dataset_id}

  eval:
    methods:
      evaluate_rows: post /v1/eval/benchmarks/{benchmark_id}/evaluations
      run_eval: post /v1/eval/benchmarks/{benchmark_id}/jobs
      evaluate_rows_alpha: post /v1/eval/benchmarks/{benchmark_id}/evaluations
      run_eval_alpha: post /v1/eval/benchmarks/{benchmark_id}/jobs

    subresources:
      jobs:
        methods:
          cancel: delete /v1/eval/benchmarks/{benchmark_id}/jobs/{job_id}
          status: get /v1/eval/benchmarks/{benchmark_id}/jobs/{job_id}
          retrieve: get /v1/eval/benchmarks/{benchmark_id}/jobs/{job_id}/result
    models:
      evaluate_response: EvaluateResponse
      benchmark_config: BenchmarkConfig
      eval_candidate: EvalCandidate
      job: Job

  inspect:
    models:
      healthInfo: HealthInfo
      providerInfo: ProviderInfo
      routeInfo: RouteInfo
      versionInfo: VersionInfo
    methods:
      health: get /v1/health
      version: get /v1/version

  inference:
    models:
      completionResponse: CompletionResponse
      embeddingsResponse: EmbeddingsResponse
      token_log_probs: TokenLogProbs
      chat_completion_response_stream_chunk: ChatCompletionResponseStreamChunk
    methods:
      chat_completion:
        deprecated: "/v1/inference/chat-completion is deprecated. Please use /v1/openai/v1/chat/completions."
        type: http
        endpoint: post /v1/inference/chat-completion
        streaming:
          stream_event_model: inference.chat_completion_response_stream_chunk
          param_discriminator: stream
      completion:
        deprecated: "/v1/inference/completion is deprecated. Please use /v1/openai/v1/completions."
        type: http
        endpoint: post /v1/inference/completion
        streaming:
          param_discriminator: stream
      batch_completion: post /v1/inference/batch-completion
      batch_chat_completion: post /v1/inference/batch-chat-completion
      embeddings:
        deprecated: "/v1/inference/embeddings is deprecated. Please use /v1/openai/v1/embeddings."
        type: http
        endpoint: post /v1/inference/embeddings
      rerank: post /v1/inference/rerank

  embeddings:
    models:
      create_embeddings_response: OpenAIEmbeddingsResponse
    methods:
      create: post /v1/openai/v1/embeddings

  chat:
    models:
      chat_completion_chunk: OpenAIChatCompletionChunk
    subresources:
      completions:
        methods:
          create:
            type: http
            endpoint: post /v1/openai/v1/chat/completions
            streaming:
              stream_event_model: chat.chat_completion_chunk
              param_discriminator: stream
          list:
            type: http
            endpoint: get /v1/openai/v1/chat/completions
          retrieve:
            type: http
            endpoint: get /v1/openai/v1/chat/completions/{completion_id}
  completions:
    methods:
      create:
        type: http
        endpoint: post /v1/openai/v1/completions
        streaming:
          param_discriminator: stream

  vector_io:
    models:
      queryChunksResponse: QueryChunksResponse
    methods:
      insert: post /v1/vector-io/insert
      query: post /v1/vector-io/query

  vector_dbs:
    models:
      list_vector_dbs_response: ListVectorDBsResponse
    methods:
      retrieve: get /v1/vector-dbs/{vector_db_id}
      list:
        endpoint: get /v1/vector-dbs
        paginated: false
      register: post /v1/vector-dbs
      unregister: delete /v1/vector-dbs/{vector_db_id}

  vector_stores:
    models:
      vector_store: VectorStoreObject
      list_vector_stores_response: VectorStoreListResponse
      vector_store_delete_response: VectorStoreDeleteResponse
      vector_store_search_response: VectorStoreSearchResponsePage
    methods:
      create: post /v1/openai/v1/vector_stores
      list:
        endpoint: get /v1/openai/v1/vector_stores
      retrieve: get /v1/openai/v1/vector_stores/{vector_store_id}
      update: post /v1/openai/v1/vector_stores/{vector_store_id}
      delete: delete /v1/openai/v1/vector_stores/{vector_store_id}
      search: post /v1/openai/v1/vector_stores/{vector_store_id}/search
    subresources:
      files:
        models:
          vector_store_file: VectorStoreFileObject
        methods:
          list: get /v1/openai/v1/vector_stores/{vector_store_id}/files
          retrieve: get /v1/openai/v1/vector_stores/{vector_store_id}/files/{file_id}
          update: post /v1/openai/v1/vector_stores/{vector_store_id}/files/{file_id}
          delete: delete /v1/openai/v1/vector_stores/{vector_store_id}/files/{file_id}
          create: post /v1/openai/v1/vector_stores/{vector_store_id}/files
          content: get /v1/openai/v1/vector_stores/{vector_store_id}/files/{file_id}/content

  models:
    models:
      model: Model
      list_models_response: ListModelsResponse
    methods:
      retrieve: get /v1/models/{model_id}
      list:
        endpoint: get /v1/models
        paginated: false
      register: post /v1/models
      unregister: delete /v1/models/{model_id}
    subresources:
      openai:
        methods:
          list:
            endpoint: get /v1/openai/v1/models
            paginated: false

  post_training:
    models:
      algorithm_config: AlgorithmConfig
      post_training_job: PostTrainingJob
      list_post_training_jobs_response: ListPostTrainingJobsResponse
    methods:
      preference_optimize: post /v1/post-training/preference-optimize
      supervised_fine_tune: post /v1/post-training/supervised-fine-tune
    subresources:
      job:
        methods:
          artifacts: get /v1/post-training/job/artifacts
          cancel: post /v1/post-training/job/cancel
          status: get /v1/post-training/job/status
          list:
            endpoint: get /v1/post-training/jobs
            paginated: false

  providers:
    models:
      list_providers_response: ListProvidersResponse
    methods:
      list:
        endpoint: get /v1/providers
        paginated: false
      retrieve: get /v1/providers/{provider_id}

  routes:
    models:
      list_routes_response: ListRoutesResponse
    methods:
      list:
        endpoint: get /v1/inspect/routes
        paginated: false

  moderations:
    models:
      create_response: ModerationObject
    methods:
      create: post /v1/openai/v1/moderations

  safety:
    models:
      run_shield_response: RunShieldResponse
    methods:
      run_shield: post /v1/safety/run-shield

  shields:
    models:
      shield: Shield
      list_shields_response: ListShieldsResponse
    methods:
      retrieve: get /v1/shields/{identifier}
      list:
        endpoint: get /v1/shields
        paginated: false
      register: post /v1/shields
      delete: delete /v1/shields/{identifier}

  synthetic_data_generation:
    models:
      syntheticDataGenerationResponse: SyntheticDataGenerationResponse
    methods:
      generate: post /v1/synthetic-data-generation/generate

  telemetry:
    models:
      span_with_status: SpanWithStatus
      trace: Trace
      query_spans_response: QuerySpansResponse
      event: Event
      query_condition: QueryCondition
    methods:
      query_traces:
        endpoint: post /v1/telemetry/traces
        skip_test_reason: "unsupported query params in java / kotlin"
      get_span_tree: post /v1/telemetry/spans/{span_id}/tree
      query_spans:
        endpoint: post /v1/telemetry/spans
        skip_test_reason: "unsupported query params in java / kotlin"
      query_metrics:
        endpoint: post /v1/telemetry/metrics/{metric_name}
        skip_test_reason: "unsupported query params in java / kotlin"
      log_event: post /v1/telemetry/events
      save_spans_to_dataset: post /v1/telemetry/spans/export
      get_span: get /v1/telemetry/traces/{trace_id}/spans/{span_id}
      get_trace: get /v1/telemetry/traces/{trace_id}

  scoring:
    methods:
      score: post /v1/scoring/score
      score_batch: post /v1/scoring/score-batch
  scoring_functions:
    methods:
      retrieve: get /v1/scoring-functions/{scoring_fn_id}
      list:
        endpoint: get /v1/scoring-functions
        paginated: false
      register: post /v1/scoring-functions
    models:
      scoring_fn: ScoringFn
      scoring_fn_params: ScoringFnParams
      list_scoring_functions_response: ListScoringFunctionsResponse

  benchmarks:
    methods:
      retrieve: get /v1/eval/benchmarks/{benchmark_id}
      list:
        endpoint: get /v1/eval/benchmarks
        paginated: false
      register: post /v1/eval/benchmarks
    models:
      benchmark: Benchmark
      list_benchmarks_response: ListBenchmarksResponse

  files:
    methods:
      create: post /v1/openai/v1/files
      list: get /v1/openai/v1/files
      retrieve: get /v1/openai/v1/files/{file_id}
      delete: delete /v1/openai/v1/files/{file_id}
      content: get /v1/openai/v1/files/{file_id}/content
    models:
      file: OpenAIFileObject
      list_files_response: ListOpenAIFileResponse
      delete_file_response: OpenAIFileDeleteResponse

settings:
  license: MIT
  unwrap_response_fields: [data]

openapi:
  transformations:
    - command: renameValue
      reason: pydantic reserved name
      args:
        filter:
          only:
            - "$.components.schemas.InferenceStep.properties.model_response"
        rename:
          python:
            property_name: "inference_model_response"

    # - command: renameValue
    #   reason: pydantic reserved name
    #   args:
    #     filter:
    #       only:
    #         - '$.components.schemas.Model.properties.model_type'
    #     rename:
    #       python:
    #         property_name: 'type'
    - command: mergeObject
      reason: Better return_type using enum
      args:
        target:
          - "$.components.schemas"
        object:
          ReturnType:
            additionalProperties: false
            properties:
              type:
                enum:
                  - string
                  - number
                  - boolean
                  - array
                  - object
                  - json
                  - union
                  - chat_completion_input
                  - completion_input
                  - agent_turn_input
            required:
              - type
            type: object
    - command: replaceProperties
      reason: Replace return type properties with better model (see above)
      args:
        filter:
          only:
            - "$.components.schemas.ScoringFn.properties.return_type"
            - "$.components.schemas.RegisterScoringFunctionRequest.properties.return_type"
        value:
          $ref: "#/components/schemas/ReturnType"
    - command: oneOfToAnyOf
      reason: Prism (mock server) doesn't like one of our requests as it technically matches multiple variants
    - reason: For better names
      command: extractToRefs
      args:
        ref:
          target: "$.components.schemas.ToolCallDelta.properties.tool_call"
          name: "#/components/schemas/ToolCallOrString"

# `readme` is used to configure the code snippets that will be rendered in the
# README.md of various SDKs. In particular, you can change the `headline`
# snippet's endpoint and the arguments to call it with.
readme:
  example_requests:
    default:
      type: request
      endpoint: post /v1/inference/chat-completion
      params: &ref_0 {}
    headline:
      type: request
      endpoint: post /v1/models
      params: *ref_0
    pagination:
      type: request
      endpoint: post /v1/inference/chat-completion
      params: {}