> ## Documentation Index > Fetch the complete documentation index at: https://docs.xpertai.cn/llms.txt > Use this file to discover all available pages before exploring further. # Start a streaming run > Start a new run and stream its output. Save run_id from the Content-Location response header. For reconnection use `GET /api/ai/threads/{thread_id}/runs/{run_id}/stream`; repeating this POST can execute the task again. The body uses assistant_id and input, not the legacy request/options envelope. ## OpenAPI ````yaml /api/openapi.sdk.json post /api/ai/threads/{thread_id}/runs/stream openapi: 3.0.3 info: title: XpertAI SDK HTTP API version: 1.0.0 description: >- Current HTTP endpoints used by the SDK guide. Authenticate with a server-side API Key using Authorization: Bearer or x-api-key. These endpoints do not require ChatKit. The reference covers the guide examples, not every method exported by the SDK. Check deployment support before using additional fields. servers: - url: https://api.xpertai.cn description: XpertAI API. Replace with your own API origin for private deployments. security: - ApiKeyBearer: [] - ApiKeyHeader: [] tags: - name: Assistants - name: Threads - name: Runs - name: Contexts - name: Store paths: /api/ai/threads/{thread_id}/runs/stream: post: tags: - Runs summary: Start a streaming run description: >- Start a new run and stream its output. Save run_id from the Content-Location response header. For reconnection use `GET /api/ai/threads/{thread_id}/runs/{run_id}/stream`; repeating this POST can execute the task again. The body uses assistant_id and input, not the legacy request/options envelope. operationId: stream_run parameters: - name: thread_id in: path required: true schema: type: string requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/RunRequest' example: assistant_id: 22222222-2222-4222-8222-222222222222 input: action: send message: input: input: Hello, introduce yourself briefly. responses: '201': description: >- Schematic SSE frames; real events contain additional fields. Preserve each id cursor, ignore colon-prefixed heartbeat comments and inspect lifecycle, interruption and error events. HTTP success or connection close alone is not task success. content: text/event-stream: schema: type: string example: >+ id: 1791763200000-0 data: {"type":"message","data":"Hello"} id: 1791763200001-0 data: {"type":"event","event":"on_conversation_end","data":{"status":"idle","error":null}} headers: Content-Location: description: Run resource path; the final path segment is run_id. schema: type: string example: >- /api/ai/threads/11111111-1111-4111-8111-111111111111/runs/33333333-3333-4333-8333-333333333333 '400': description: Invalid input, identifier mismatch or invalid operation state. '401': description: Missing or invalid API Key. '403': description: The caller cannot access the requested resource. '404': description: Resource not found or unavailable to the caller. components: schemas: RunRequest: type: object properties: assistant_id: type: string description: >- Published agent ID. HTTP uses assistant_id; the SDK accepts assistantId as a method argument. input: oneOf: - $ref: '#/components/schemas/SendInput' - $ref: '#/components/schemas/ResumeInput' - $ref: '#/components/schemas/RetryInput' - $ref: '#/components/schemas/FollowUpInput' discriminator: propertyName: action mapping: send: '#/components/schemas/SendInput' retry: '#/components/schemas/RetryInput' resume: '#/components/schemas/ResumeInput' follow_up: '#/components/schemas/FollowUpInput' description: >- Chat action payload. Examples show send; retry, resume and follow_up require the appropriate execution state and access. context: type: object additionalProperties: true description: >- Optional runtime context. Use only fields supported by your deployed agent. required: - assistant_id - input SendInput: type: object properties: action: type: string enum: - send conversationId: type: string description: >- If supplied, must match the conversation owning the URL thread. Read metadata.conversation_id from the thread response. message: $ref: '#/components/schemas/Message' state: type: object additionalProperties: true projectId: type: string environmentId: type: string sandboxEnvironmentId: type: string required: - action - message ResumeInput: type: object properties: action: type: string enum: - resume conversationId: type: string description: >- If supplied, must match the conversation owning the URL thread. Read metadata.conversation_id from the thread response. target: $ref: '#/components/schemas/Target' decision: type: object properties: type: type: string enum: - confirm - reject payload: {} required: - type patch: type: object properties: agentKey: type: string toolCalls: type: array items: {} update: {} state: type: object additionalProperties: true required: - action - target - decision RetryInput: type: object properties: action: type: string enum: - retry conversationId: type: string description: >- If supplied, must match the conversation owning the URL thread. Read metadata.conversation_id from the thread response. source: $ref: '#/components/schemas/Target' environmentId: type: string sandboxEnvironmentId: type: string required: - action - source FollowUpInput: type: object properties: action: type: string enum: - follow_up conversationId: type: string description: >- If supplied, must match the conversation owning the URL thread. Read metadata.conversation_id from the thread response. mode: type: string enum: - queue - steer message: $ref: '#/components/schemas/Message' target: $ref: '#/components/schemas/Target' state: type: object additionalProperties: true required: - action - mode - message Message: type: object properties: clientMessageId: type: string description: >- Caller message identifier; it is not a server-side idempotency guarantee. input: type: object properties: input: type: string description: User text. Additional inputs depend on the agent definition. files: type: array items: $ref: '#/components/schemas/AgentFile' additionalProperties: true required: - input Target: type: object properties: aiMessageId: type: string executionId: type: string description: >- Use identifiers returned by the original execution or interruption; do not invent target IDs. AgentFile: type: object properties: id: type: string description: FileAsset ID. Pass the returned handle in message.input.files. fileId: type: string storageFileId: type: string description: >- StorageFile ID for knowledgebase document associations; do not substitute the FileAsset ID. url: type: string originalName: type: string size: type: integer mimeType: type: string status: type: string parseStatus: type: string parseMode: type: string purpose: type: string required: - id securitySchemes: ApiKeyBearer: type: http scheme: bearer description: Server-side API Key; do not expose it in browser code. ApiKeyHeader: type: apiKey in: header name: x-api-key ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.