Class CompletionResource

constructor

• new CompletionResource(http: HttpClient): CompletionResource

  Parameters

  • http: HttpClient

  Returns CompletionResource

Readonlystream

create

• create(
      input: {
          attribution?: Record<string, unknown>;
          context?: Record<string, unknown>;
          messages: {
              content?: string;
              content_parts?: Record<string, unknown>[];
              resume_token?: string;
              role: string;
              structured_output?: unknown;
              tool_calls?: {
                  arguments: Record<string, unknown>;
                  id: string;
                  name: string;
                  thought_signature?: string;
              }[];
              tool_results?: {
                  content?: string;
                  id: string;
                  name: string;
                  resolution?: unknown;
              }[];
          }[];
          opts: {
              max_tokens?: number;
              model: string;
              server_tools?: Record<string, unknown>[];
              structured_output?: Record<string, unknown>;
              temperature?: number;
              tool_choice?: string;
              tools?: {
                  function: {
                      description?: string;
                      name: string;
                      parameters: Record<string, unknown>;
                  };
                  type: string;
              }[];
          };
          session_id?: string;
      },
  ): Promise<AICompletionResult>

  Create a chat completion Sends a list of messages to the configured AI provider and returns a single completion. Use this endpoint when you want direct, low-level access to the underlying model without any workflow or agent orchestration. The authenticated app must have the llm_calls entitlement enabled on its plan. Requests that exceed the plan quota are rejected with 402. Token usage is recorded against the authenticated app and organization. Supply tools and tool_choice to enable OpenAI-compatible function calling. Use server_tools to activate platform-managed tools such as search that run on the server side before the response is returned.

  Parameters

  • input: {
        attribution?: Record<string, unknown>;
        context?: Record<string, unknown>;
        messages: {
            content?: string;
            content_parts?: Record<string, unknown>[];
            resume_token?: string;
            role: string;
            structured_output?: unknown;
            tool_calls?: {
                arguments: Record<string, unknown>;
                id: string;
                name: string;
                thought_signature?: string;
            }[];
            tool_results?: {
                content?: string;
                id: string;
                name: string;
                resolution?: unknown;
            }[];
        }[];
        opts: {
            max_tokens?: number;
            model: string;
            server_tools?: Record<string, unknown>[];
            structured_output?: Record<string, unknown>;
            temperature?: number;
            tool_choice?: string;
            tools?: {
                function: {
                    description?: string;
                    name: string;
                    parameters: Record<string, unknown>;
                };
                type: string;
            }[];
        };
        session_id?: string;
    }

    Request body.

    • Optionalattribution?: Record<string, unknown>

      Optional caller tags recorded on the LLM-call accounting row. Allowed keys: task_id (max 64 chars), work_id (max 64), agent_kind (one of worker, one_off, subagent, overseer, task_reviewer, interactive, daemon_job), agent_id (max 64), client (max 32). Anything else is a 400.

    • Optionalcontext?: Record<string, unknown>

      Key-value map used to resolve template variables in message content. Omit if messages contain no templates.

    • messages: {
          content?: string;
          content_parts?: Record<string, unknown>[];
          resume_token?: string;
          role: string;
          structured_output?: unknown;
          tool_calls?: {
              arguments: Record<string, unknown>;
              id: string;
              name: string;
              thought_signature?: string;
          }[];
          tool_results?: {
              content?: string;
              id: string;
              name: string;
              resolution?: unknown;
          }[];
      }[]

      Ordered list of conversation messages to send to the model.

    • opts: {
          max_tokens?: number;
          model: string;
          server_tools?: Record<string, unknown>[];
          structured_output?: Record<string, unknown>;
          temperature?: number;
          tool_choice?: string;
          tools?: {
              function: {
                  description?: string;
                  name: string;
                  parameters: Record<string, unknown>;
              };
              type: string;
          }[];
      }

      Model and sampling configuration for this request.

    • Optionalsession_id?: string

      Optional UUID grouping this and other completions under one session in the Developers dashboard. Pass the same value across requests to link them; omit to auto-generate a per-request session.

  Returns Promise<AICompletionResult>

  The completed AI response, including the generated message, finish reason, and token usage.