### Start ThinkParse with Docker Compose Source: https://github.com/wzdavid/thinkparse/blob/main/docs/quickstart.zh.md Validates the compose configuration and starts the services in detached mode with build. The first build downloads models and may take a long time; `up` waits for MinerU health (start_period up to ~600s) before starting the gateway. ```bash docker compose --env-file .env -f docker/docker-compose.yml config --services docker compose --env-file .env -f docker/docker-compose.yml up -d --build ``` -------------------------------- ### GET /api/v1/queue/tasks Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取队列中的任务列表。 ```APIDOC ## GET /api/v1/queue/tasks ### Description 获取队列中的任务列表。返回 tasks[] 数组，包含 task_id、status、file_name。 ### Method GET ### Endpoint /api/v1/queue/tasks ### Response #### Success Response (200) - **tasks** (array) - 任务列表 - **task_id** (string) - 任务 ID - **status** (string) - 状态 - **file_name** (string) - 文件名 ``` -------------------------------- ### GET /api/v1/health/ready Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 检查服务是否就绪。 ```APIDOC ## GET /api/v1/health/ready ### Description 检查服务是否就绪。任务库、对象存储、至少一个 MinerU 健康时返回 200，否则 503。 ### Method GET ### Endpoint /api/v1/health/ready ### Response #### Success Response (200) - 无特定字段 #### Error Responses - **503** - 服务未就绪 ``` -------------------------------- ### GET /api/v1/health/live Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 检查网关进程存活。 ```APIDOC ## GET /api/v1/health/live ### Description 检查网关进程是否存活。 ### Method GET ### Endpoint /api/v1/health/live ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v2/tiers Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Lists the processing tiers available in this installation. Does not expose MinerU model repository names. ```APIDOC ## GET /api/v2/tiers ### Description Lists the processing tiers available in this installation. Does not expose MinerU model repository names. ### Method GET ### Endpoint /api/v2/tiers ### Response #### Success Response (200) - **data** (array) - List of tier objects. - **id** (string) - Tier identifier. - **discovered** (boolean) - Whether tiers were discovered dynamically. #### Response Example { "data": [ { "id": "standard" } ], "discovered": true } ``` -------------------------------- ### GET /api/v2/jobs Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取任务列表。 ```APIDOC ## GET /api/v2/jobs ### Description 获取任务列表，支持按状态和数量过滤。 ### Method GET ### Endpoint /api/v2/jobs ### Parameters #### Query Parameters - **status** (string) - Optional - 任务状态 - **limit** (integer) - Optional - 返回数量限制 ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v1/queue/stats Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取队列统计信息。 ```APIDOC ## GET /api/v1/queue/stats ### Description 获取队列统计信息。返回 stats.pending、stats.processing 等字段。 ### Method GET ### Endpoint /api/v1/queue/stats ### Response #### Success Response (200) - **stats** (object) - 队列统计 - **pending** (integer) - 等待中的任务数 - **processing** (integer) - 处理中的任务数 ``` -------------------------------- ### Two-GPU .env configuration Source: https://github.com/wzdavid/thinkparse/blob/main/docs/DEPLOYMENT.zh.md Example .env settings for a two-GPU deployment. THINKPARSE_SLOTS must be a comma-separated list with one entry per MinerU address, each equal to MINERU_GPU_CONCURRENCY. ```bash COMPOSE_PROFILES=gpu,gpu1 MINERU_BASE_URLS=http://mineru-gpu0:8002,http://mineru-gpu1:8002 MINERU_GPU_CONCURRENCY=2 THINKPARSE_SLOTS=2,2 ``` -------------------------------- ### GET /api/v2/jobs/{id} Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取任务详情。 ```APIDOC ## GET /api/v2/jobs/{id} ### Description 获取任务详情，包含 timing、attempt、pages、batch_id、priority 等字段。 ### Method GET ### Endpoint /api/v2/jobs/{id} ### Parameters #### Path Parameters - **id** (string) - Required - 任务 ID ### Response #### Success Response (200) - **timing** (object) - 时间统计 - **queue_ms** (integer) - 排队耗时（毫秒） - **parse_ms** (integer) - 解析耗时（毫秒） - **project_ms** (integer) - 投影耗时（毫秒） - **attempt** (integer) - 尝试次数 - **pages** (integer) - 页数 - **batch_id** (string) - 批次 ID - **priority** (integer) - 优先级 ``` -------------------------------- ### GET /api/v2/tiers Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取可用档位列表。 ```APIDOC ## GET /api/v2/tiers ### Description 获取可用档位列表。返回 data[].id 与 discovered。 ### Method GET ### Endpoint /api/v2/tiers ### Response #### Success Response (200) - **data** (array) - 档位列表 - **id** (string) - 档位 ID - **discovered** (boolean) - 是否发现 ``` -------------------------------- ### GET /api/v2/files/{id} Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取文件信息。 ```APIDOC ## GET /api/v2/files/{id} ### Description 获取文件信息。 ### Method GET ### Endpoint /api/v2/files/{id} ### Parameters #### Path Parameters - **id** (string) - Required - 文件 ID ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v1/health/deep Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取上游详情、在途等运维信息。 ```APIDOC ## GET /api/v1/health/deep ### Description 获取上游详情、在途等运维信息。 ### Method GET ### Endpoint /api/v1/health/deep ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v2/batches/{id} Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取批次信息。 ```APIDOC ## GET /api/v2/batches/{id} ### Description 获取批次信息。 ### Method GET ### Endpoint /api/v2/batches/{id} ### Parameters #### Path Parameters - **id** (string) - Required - 批次 ID ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v1/tasks/{task_id} Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 查询任务状态，返回任务详情和完成时的解析结果。 ```APIDOC ## GET /api/v1/tasks/{task_id} ### Description 查询任务状态。返回任务详情，包括状态、文件名、后端等。任务完成时顶层包含 markdown_content、content_list、middle_json、images 等字段。 ### Method GET ### Endpoint /api/v1/tasks/{task_id} ### Parameters #### Path Parameters - **task_id** (string) - Required - 任务 ID ### Response #### Success Response (200) - **success** (boolean) - 固定为 true - **task** (object) - 任务详情 - **task_id** (string) - 任务 ID - **status** (string) - 状态：pending、processing、completed、failed、cancelled - **file_name** (string) - 文件名 - **backend** (string) - 后端标识 - **error_message** (string|null) - 错误信息，失败时非空 - **started_at** (string|null) - 开始时间 - **completed_at** (string|null) - 完成时间 - **markdown_content** (string) - 完成时存在，Markdown 内容 - **content_list** (array) - 完成时存在，内容列表 - **middle_json** (object) - 完成时存在，中间 JSON - **images** (array) - 完成时存在，图片数组 - **data.content** (string) - 完成时存在，与 markdown_content 相同 #### Response Example { "success": true, "task": { "task_id": "...", "status": "pending", "file_name": "paper.pdf", "backend": "pipeline", "error_message": null, "started_at": null, "completed_at": null } } #### Error Responses - **404** - 未知任务 ``` -------------------------------- ### GET /api/v2/batches/{id}/jobs Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取批次下的任务列表。 ```APIDOC ## GET /api/v2/batches/{id}/jobs ### Description 获取批次下的任务列表。 ### Method GET ### Endpoint /api/v2/batches/{id}/jobs ### Parameters #### Path Parameters - **id** (string) - Required - 批次 ID ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v2/stats Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取统计信息。 ```APIDOC ## GET /api/v2/stats ### Description 获取统计信息，包括队列、槽位、在途字节、时间窗内完成数等。通常需要 API key。 ### Method GET ### Endpoint /api/v2/stats ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v2/health Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 检查服务健康状态。 ```APIDOC ## GET /api/v2/health ### Description 检查服务健康状态。任务库、对象存储、每个上游健康与档位；任一 MinerU 可达即 200。可不带 key，适合负载均衡探活。 ### Method GET ### Endpoint /api/v2/health ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v2/uploads/{id} Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取上传会话信息。 ```APIDOC ## GET /api/v2/uploads/{id} ### Description 获取上传会话信息。 ### Method GET ### Endpoint /api/v2/uploads/{id} ### Parameters #### Path Parameters - **id** (string) - Required - 上传会话 ID ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v1/health Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md General health endpoint. Behavior not explicitly detailed in source. ```APIDOC ## GET /api/v1/health ### Description General health endpoint. Behavior not explicitly detailed in source. ### Method GET ### Endpoint /api/v1/health ### Parameters None ### Request Example (No explicit example provided in source) ### Response #### Success Response (200) - (No specific fields documented) #### Response Example (No explicit example provided in source) ``` -------------------------------- ### GET /api/v2/files/{id}/content Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 获取文件内容。 ```APIDOC ## GET /api/v2/files/{id}/content ### Description 获取文件内容。 ### Method GET ### Endpoint /api/v2/files/{id}/content ### Parameters #### Path Parameters - **id** (string) - Required - 文件 ID ### Response #### Success Response (200) - 无特定字段 ``` -------------------------------- ### GET /api/v1/health/live Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Health check that only indicates the gateway process is alive. ```APIDOC ## GET /api/v1/health/live ### Description Health check that only indicates the gateway process is alive. ### Method GET ### Endpoint /api/v1/health/live ### Parameters None ### Request Example (No explicit example provided in source) ### Response #### Success Response (200) - (No specific fields documented) #### Response Example (No explicit example provided in source) ``` -------------------------------- ### GET /api/v1/health/ready Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Health check that returns 200 when task database, object storage, and MinerU /v1/health are all available; otherwise 503. ```APIDOC ## GET /api/v1/health/ready ### Description Health check that returns 200 when task database, object storage, and MinerU /v1/health are all available; otherwise 503. ### Method GET ### Endpoint /api/v1/health/ready ### Parameters None ### Request Example (No explicit example provided in source) ### Response #### Success Response (200) - (No specific fields documented) #### Error Responses - **503** - Service unavailable. #### Response Example (No explicit example provided in source) ``` -------------------------------- ### GET /api/v2/stats Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Returns current queue length, slots, in-flight bytes, and completion count for a recent time window. Requires API key. ```APIDOC ## GET /api/v2/stats ### Description Returns current queue length, slots, in-flight bytes, and completion count for a recent time window. Requires API key. ### Method GET ### Endpoint /api/v2/stats ### Authentication - **Authorization** (header) - Required - Bearer token. ### Response #### Success Response (200) - **queue** (integer) - Current queue length. - **slots** (integer) - Available slots. - **in_flight_bytes** (integer) - Bytes currently in flight. - **completed_recent** (integer) - Number of completions in the recent time window. #### Response Example { "queue": 5, "slots": 10, "in_flight_bytes": 1024, "completed_recent": 100 } ``` -------------------------------- ### POST /file_parse Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Legacy endpoint that creates a task and waits for its completion, then returns the same JSON as GET /api/v1/tasks/{id}. New integrations should not use this endpoint. ```APIDOC ## POST /file_parse ### Description Legacy endpoint that creates a task and waits for its completion, then returns the same JSON as GET /api/v1/tasks/{id}. New integrations should not use this endpoint. ### Method POST ### Endpoint /file_parse ### Parameters #### Request Body (multipart/form-data) - **file** (file) - Required - The file to be parsed. - Other form fields as described in section 5 (legacy mapping). ### Request Example (No explicit example provided in source) ### Response #### Success Response (200) Same JSON structure as GET /api/v1/tasks/{id}. #### Response Example (No explicit example provided in source) ``` -------------------------------- ### GET /api/v2/jobs/{id} Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Retrieves the status and details of a specific job, including timing, attempt, pages, and batch_id. ```APIDOC ## GET /api/v2/jobs/{id} ### Description Retrieves the status and details of a specific job, including timing, attempt, pages, and batch_id. ### Method GET ### Endpoint /api/v2/jobs/{id} ### Parameters #### Path Parameters - **id** (string) - Required - The job identifier. ### Response #### Success Response (200) - **id** (string) - Job identifier. - **status** (string) - Job status. - **timing** (object) - Timing details. - **queue_ms** (integer) - Time spent in queue. - **parse_ms** (integer) - Time spent parsing. - **project_ms** (integer) - Time spent projecting. - **attempt** (integer) - Current attempt number. - **pages** (integer) - Number of pages processed. - **batch_id** (string) - Associated batch identifier. #### Response Example { "id": "job-id", "status": "completed", "timing": { "queue_ms": 100, "parse_ms": 500, "project_ms": 200 }, "attempt": 1, "pages": 10, "batch_id": "batch-id" } ``` -------------------------------- ### GET /api/v1/queue/stats Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Returns queue statistics. Pending and processing counts come from the task database; completed and failed counts are real counts. ```APIDOC ## GET /api/v1/queue/stats ### Description Returns queue statistics. Pending and processing counts come from the task database; completed and failed counts are real counts. ### Method GET ### Endpoint /api/v1/queue/stats ### Parameters None ### Request Example (No explicit example provided in source) ### Response #### Success Response (200) - **stats** (object) - Contains: - **pending** (integer) - Number of pending tasks. - **processing** (integer) - Number of processing tasks. - **completed** (integer) - Number of completed tasks. - **failed** (integer) - Number of failed tasks. #### Response Example (No explicit example provided in source) ``` -------------------------------- ### GET /api/v1/tasks/{task_id} Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Retrieves the status and result of a task. Returns a task object with status and, when completed, the parsed content fields. ```APIDOC ## GET /api/v1/tasks/{task_id} ### Description Retrieves the status and result of a task. Returns a task object with status and, when completed, the parsed content fields. ### Method GET ### Endpoint /api/v1/tasks/{task_id} ### Parameters #### Path Parameters - **task_id** (string) - Required - The ID of the task. ### Request Example (No explicit example provided in source) ### Response #### Success Response (200) - **success** (boolean) - Always true. - **task** (object) - Contains task details. - **task_id** (string) - Task ID. - **status** (string) - One of: pending, processing, completed, failed, cancelled. - **file_name** (string) - Name of the file. - **backend** (string) - Backend used. - **error_message** (string|null) - Error message if failed, otherwise null. - **started_at** (string|null) - Start timestamp. - **completed_at** (string|null) - Completion timestamp. - When status is completed, additional top-level fields: - **markdown_content** (string) - Markdown content with relative image paths. - **content_list** (array) - Array of objects with fields: type, text, page_idx, text_level, bbox, table and image fields. - **middle_json** (object) - Contains pdf_info array; each page has page_size and discarded_blocks. - **images** (array) - Array of image filenames; empty array if no images. - **data.content** (string) - Same as markdown_content. #### Response Example (In Progress) ```json { "success": true, "task": { "task_id": "...", "status": "pending", "file_name": "paper.pdf", "backend": "pipeline", "error_message": null, "started_at": null, "completed_at": null } } ``` #### Error Responses - **404** - Unknown task. ``` -------------------------------- ### GET /api/v2/health Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Returns health status of the task database, object storage, each MinerU instance, current tiers, and in-flight count. Returns 200 if any MinerU is reachable. Does not require API key, for load balancer health checks. ```APIDOC ## GET /api/v2/health ### Description Returns health status of the task database, object storage, each MinerU instance, current tiers, and in-flight count. Returns 200 if any MinerU is reachable. Does not require API key, for load balancer health checks. ### Method GET ### Endpoint /api/v2/health ### Response #### Success Response (200) - **database** (string) - Health status of the task database. - **storage** (string) - Health status of object storage. - **mineru** (array) - Health status of each MinerU instance. - **tiers** (array) - Current available tiers. - **in_flight** (integer) - Number of in-flight jobs. #### Response Example { "database": "ok", "storage": "ok", "mineru": ["ok"], "tiers": ["standard"], "in_flight": 3 } ``` -------------------------------- ### GET /api/v1/health/deep Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Health check that provides per-GPU worker details, in-flight counts, and last projection failure. For operational use only. ```APIDOC ## GET /api/v1/health/deep ### Description Health check that provides per-GPU worker details, in-flight counts, and last projection failure. For operational use only. ### Method GET ### Endpoint /api/v1/health/deep ### Parameters None ### Request Example (No explicit example provided in source) ### Response #### Success Response (200) - (No specific fields documented) #### Response Example (No explicit example provided in source) ``` -------------------------------- ### Clone repository and copy .env.example Source: https://github.com/wzdavid/thinkparse/blob/main/docs/quickstart.zh.md Clones the ThinkParse repository, enters the directory, and copies the environment template to .env. Required before editing configuration. ```bash git clone https://github.com/wzdavid/ThinkParse.git cd ThinkParse cp .env.example .env ``` -------------------------------- ### 快速开始：克隆、配置与启动 Source: https://github.com/wzdavid/thinkparse/blob/main/README.zh.md 快速启动 ThinkParse 的完整命令序列：克隆仓库、配置环境变量、按需选择 COMPOSE_PROFILES（gpu 或 cpu），启动容器并验证健康检查与档位接口。 ```bash git clone https://github.com/wzdavid/ThinkParse.git cd ThinkParse cp .env.example .env # 三选一：COMPOSE_PROFILES=gpu | cpu | （留空接外部 MinerU） # 修改 POSTGRES_PASSWORD、MINIO_ROOT_USER、MINIO_ROOT_PASSWORD docker compose --env-file .env -f docker/docker-compose.yml up -d --build curl -fsS http://127.0.0.1:8000/api/v1/health/live curl -sS http://127.0.0.1:8000/api/v1/health/ready curl -sS http://127.0.0.1:8000/api/v2/tiers ``` -------------------------------- ### 验证命令（unittest） Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0_PLAN.zh.md 运行控制面测试的命令。 ```bash cd /Users/david/Blue/ThinkParse PYTHONPATH=. python -m unittest discover -s tests -p 'test_control_*.py' ``` -------------------------------- ### Record resource baseline with nvidia-smi, free, and /api/v1/health/deep Source: https://github.com/wzdavid/thinkparse/blob/main/docs/DEPLOYMENT.zh.md Record resource baseline while one task is in flight, including GPU memory and utilization, free memory, and the deep health endpoint. ```bash nvidia-smi --query-gpu=index,memory.used,utilization.gpu --format=csv free -h curl -sS http://127.0.0.1:8000/api/v1/health/deep ``` -------------------------------- ### Verify health and tiers endpoints Source: https://github.com/wzdavid/thinkparse/blob/main/docs/quickstart.zh.md Checks service readiness: live returns HTTP 200, ready returns 200 when MinerU is reachable with components.task_store and components.object_store true, and tiers shows discovered: true with a list matching the selected mode. ```bash curl -fsS http://127.0.0.1:8000/api/v1/health/live curl -sS http://127.0.0.1:8000/api/v1/health/ready curl -sS http://127.0.0.1:8000/api/v2/tiers ``` -------------------------------- ### 开发：安装依赖、运行测试与检查 Source: https://github.com/wzdavid/thinkparse/blob/main/README.zh.md 开发环境安装依赖并运行测试与代码检查。 ```bash pip install -r control/requirements-dev.txt PYTHONPATH=. python -m unittest discover -s tests -p 'test_control_*.py' ruff check control tests ``` -------------------------------- ### 查看日志并停止服务 Source: https://github.com/wzdavid/thinkparse/blob/main/docs/DEPLOYMENT.zh.md 查看 gateway、reconciler 和 mineru-gpu 的实时日志，然后停止服务但保留卷；使用 down -v 会删除任务与产物。 ```bash docker compose --env-file .env -f docker/docker-compose.yml logs -f gateway reconciler mineru-gpu docker compose --env-file .env -f docker/docker-compose.yml down # 保留卷 # down -v 会删除任务与产物 ``` -------------------------------- ### Check services and query /api/v2/tiers Source: https://github.com/wzdavid/thinkparse/blob/main/docs/DEPLOYMENT.zh.md Check running services and query the tier endpoint. discovered: true means tiers were read from upstream; false means upstream is not reachable yet and the list only shows configured allowed values. ```bash docker compose --env-file .env -f docker/docker-compose.yml ps ``` ```bash curl -sS http://127.0.0.1:8000/api/v2/tiers ``` -------------------------------- ### 异步提交任务：curl 命令 Source: https://github.com/wzdavid/thinkparse/blob/main/README.zh.md 异步提交 PDF 文档到 /api/v1/tasks/submit，使用 pipeline 后端。适用于批量与生产场景。 ```bash curl -sS \ -F "file=@document.pdf" \ -F "backend=pipeline" \ http://127.0.0.1:8000/api/v1/tasks/submit ``` -------------------------------- ### Health check with curl for /api/v1/health/live and /api/v1/health/ready Source: https://github.com/wzdavid/thinkparse/blob/main/docs/DEPLOYMENT.zh.md Check the live and ready health endpoints. The live endpoint should return 200, and the ready endpoint returns 200 when MinerU is reachable, otherwise 503. In both cases, components.task_store and components.object_store should be true; do not submit jobs when components.mineru is false. ```bash curl -fsS http://127.0.0.1:8000/api/v1/health/live curl -sS -D - http://127.0.0.1:8000/api/v1/health/ready -o /tmp/thinkparse-ready.json cat /tmp/thinkparse-ready.json ``` -------------------------------- ### POST /api/v2/uploads Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Creates a new upload resource, registering byte count and checksum before content is uploaded. ```APIDOC ## POST /api/v2/uploads ### Description Creates a new upload resource, registering byte count and checksum before content is uploaded. ### Method POST ### Endpoint /api/v2/uploads ### Parameters #### Request Body - **byte_count** (integer) - Required - The number of bytes in the file to be uploaded. - **checksum** (string) - Required - The SHA-256 checksum of the file content. ### Request Example { "byte_count": 12345, "checksum": "sha256hex..." } ### Response #### Success Response (200) - **id** (string) - The unique identifier of the created upload resource. #### Response Example { "id": "upload-id" } ``` -------------------------------- ### Submit PDF task Source: https://github.com/wzdavid/thinkparse/blob/main/docs/quickstart.zh.md Submits a PDF file for processing using the pipeline backend, which maps to MinerU basic tier. The response contains a task_id to use for polling. ```bash curl -sS \ -F "file=@tests/files/2604.04771v2.pdf" \ -F "backend=pipeline" \ http://127.0.0.1:8000/api/v1/tasks/submit ``` -------------------------------- ### POST /api/v2/jobs Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 创建解析任务。 ```APIDOC ## POST /api/v2/jobs ### Description 创建解析任务。body 含 file_id（或 MinerU 风格的 files[].source.file_id），可选 tier、ocr_mode、batch_id、priority。 ### Method POST ### Endpoint /api/v2/jobs ### Parameters #### Request Body - **file_id** (string) - Required - 文件 ID - **files[].source.file_id** (string) - Optional - MinerU 风格的文件引用 - **tier** (string) - Optional - 档位 - **ocr_mode** (string) - Optional - OCR 模式 - **batch_id** (string) - Optional - 批次 ID，字母、数字、-、_，最长 64 - **priority** (integer) - Optional - 优先级，0-9，越大越优先 ### Response #### Success Response (202) - 无特定字段 #### Error Responses - **400** - 部署做不到的档位直接返回 400 ``` -------------------------------- ### GET /api/v2/jobs Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Lists jobs, optionally filtered by status and limited by count. ```APIDOC ## GET /api/v2/jobs ### Description Lists jobs, optionally filtered by status and limited by count. ### Method GET ### Endpoint /api/v2/jobs ### Parameters #### Query Parameters - **status** (string) - Optional - Filter by job status. - **limit** (integer) - Optional - Maximum number of jobs to return. ### Response #### Success Response (200) - **jobs** (array) - List of job objects. - **id** (string) - Job identifier. - **status** (string) - Job status. #### Response Example { "jobs": [ { "id": "job-id", "status": "queued" } ] } ``` -------------------------------- ### POST /api/v2/uploads/{id}/complete Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 完成上传会话。 ```APIDOC ## POST /api/v2/uploads/{id}/complete ### Description 完成上传会话。 ### Method POST ### Endpoint /api/v2/uploads/{id}/complete ### Parameters #### Path Parameters - **id** (string) - Required - 上传会话 ID ### Response #### Success Response (202) - 无特定字段 ``` -------------------------------- ### View gateway and reconciler logs Source: https://github.com/wzdavid/thinkparse/blob/main/docs/quickstart.zh.md Follows logs for the gateway and reconciler containers, useful for diagnosing issues like tasks stuck in processing or tiers not discovered. ```bash docker compose --env-file .env -f docker/docker-compose.yml logs -f gateway reconciler ``` -------------------------------- ### POST /api/v2/uploads Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 创建上传会话。 ```APIDOC ## POST /api/v2/uploads ### Description 创建上传会话。 ### Method POST ### Endpoint /api/v2/uploads ### Response #### Success Response (202) - 无特定字段 ``` -------------------------------- ### 代码结构（control/ 目录） Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0_PLAN.zh.md 列出 2.0 控制面的主要文件和目录结构。 ```text control/ 2.0 控制面，不加载 CUDA gateway.py /api/v1 兼容与 /api/v2 产品接口。MinerU /v1 只在协调器内部调用 reconciler.py 独立进程入口 reconcile.py 投递、轮询、重试、取消、投影、超时、清理 mineru.py MinerU 4.0 HTTP docling.py docling-serve HTTP，只接非 PDF project.py MinerU middle → 3.x 形状；Docling 保留 docling_document store.py PostgreSQL 与内存任务库 objects.py 本地目录；设置 endpoint 后走外部 S3 docker/docker-compose.yml cpu 与 gpu 两个 profile，别名都是 mineru-router docker/mineru.Dockerfile GPU，默认下载 standard docker/mineru-cpu.Dockerfile CPU，ONNX basic tests/test_control_*.py tests/fixtures/legacy_projection/ ``` -------------------------------- ### POST /api/v1/tasks/submit Source: https://github.com/wzdavid/thinkparse/blob/main/docs/api.zh.md 提交一个新的解析任务，使用 multipart/form-data 上传文件，可选 priority 参数控制优先级。 ```APIDOC ## POST /api/v1/tasks/submit ### Description 提交一个新的解析任务。请求为 multipart/form-data，文件字段名为 file。可选表单 priority 控制优先级（0-9，越大越优先，默认 0）。 ### Method POST ### Endpoint /api/v1/tasks/submit ### Parameters #### Request Body - **file** (file) - Required - 要解析的文件 - **priority** (integer) - Optional - 优先级，0-9，越大越优先，默认 0。超出范围返回 400。 ### Response #### Success Response (200) - **success** (boolean) - 固定为 true - **task_id** (string) - ThinkParse 生成的任务 ID - **status** (string) - 固定为 "pending" - **file_name** (string) - 文件名 - **created_at** (string) - 创建时间 - **backend** (string) - 后端标识 - **priority** (integer) - 优先级 #### Response Example { "success": true, "task_id": "...", "status": "pending", "file_name": "paper.pdf", "created_at": "...", "backend": "pipeline", "priority": 0 } #### Error Responses - **400** - priority 超出范围 - **413** - 文件过大 - **507** - 本地对象目录水位不足（外部 S3 不做此项检查） ``` -------------------------------- ### ThinkParse 架构图 Source: https://github.com/wzdavid/thinkparse/blob/main/README.zh.md 展示 ThinkParse 的整体架构：客户端通过 /api/v1 或 /api/v2 提交任务，API 与任务编排层处理后，通过 HTTP 将任务分发到 MinerU 4.0 或 Docling Serve。 ```text 你的应用 │ POST /api/v1/tasks/submit 或 /api/v2/uploads + /api/v2/jobs ▼ ┌──────────────────────────────────────────────┐ │ ThinkParse（API + 任务编排） │ │ 网关 · PostgreSQL · 对象存储 · 协调循环 │ └──────────────────┬───────────────────────────┘ │ HTTP ┌──────────┴──────────┐ ▼ ▼ MinerU 4.0（多 GPU） Docling Serve（可选） ``` -------------------------------- ### POST /api/v2/uploads/{id}/complete Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Marks an upload as complete, allowing a job to be created referencing the file. ```APIDOC ## POST /api/v2/uploads/{id}/complete ### Description Marks an upload as complete, allowing a job to be created referencing the file. ### Method POST ### Endpoint /api/v2/uploads/{id}/complete ### Parameters #### Path Parameters - **id** (string) - Required - The unique identifier of the upload resource. ### Response #### Success Response (200) - **status** (string) - Confirmation that the upload is complete. #### Response Example { "status": "complete" } ``` -------------------------------- ### POST /api/v2/jobs Source: https://github.com/wzdavid/thinkparse/blob/main/docs/design/THINKPARSE_2.0.zh.md Creates a new parsing job. Accepts a file_id or MinerU-style files[].source.file_id, and optionally a batch_id. Returns 202 Accepted. ```APIDOC ## POST /api/v2/jobs ### Description Creates a new parsing job. Accepts a file_id or MinerU-style files[].source.file_id, and optionally a batch_id. Returns 202 Accepted. ### Method POST ### Endpoint /api/v2/jobs ### Parameters #### Request Body - **file_id** (string) - Optional - The identifier of the uploaded file to parse. - **files** (array) - Optional - MinerU-style file references. - **source.file_id** (string) - Required - The file identifier. - **batch_id** (string) - Optional - Identifier of a batch to associate the job with. - **tier** (string) - Required - The processing tier (e.g., standard, premium). - **ocr_mode** (string) - Required - The OCR mode to use. ### Request Example { "file_id": "file-id", "batch_id": "batch-id", "tier": "standard", "ocr_mode": "auto" } ### Response #### Success Response (202) - **id** (string) - The job identifier. - **status** (string) - Initial status, e.g., "queued". #### Response Example { "id": "job-id", "status": "queued" } ``` -------------------------------- ### Save markdown content to file with python3 one-liner Source: https://github.com/wzdavid/thinkparse/blob/main/docs/DEPLOYMENT.zh.md Save the markdown_content from the task response to /tmp/thinkparse-result.md for reading instead of verification. ```bash python3 -c 'import json; from pathlib import Path; body=json.loads(Path("/tmp/thinkparse-task.json").read_text()); Path("/tmp/thinkparse-result.md").write_text(body["markdown_content"])' ``` -------------------------------- ### Summarize task response with python3 heredoc Source: https://github.com/wzdavid/thinkparse/blob/main/docs/DEPLOYMENT.zh.md Print a summary of the completed task response without opening the large file directly. It reports status, error_message, page count, page_size, schema_key, equation and table counts, markdown length and equality with data.content, data:image occurrences, image metadata, and the sample header. The script intentionally does not print data_url values. ```bash python3 - <<'PY' import json from pathlib import Path body = json.loads(Path("/tmp/thinkparse-task.json").read_text()) task = body["task"] middle = body.get("middle_json") or {} info = middle.get("pdf_info") or [] content = body.get("content_list") or [] images = body.get("images") or [] markdown = body.get("markdown_content") or "" equations = [item for item in content if item.get("type") == "equation" and item.get("text")] tables = [item for item in content if item.get("type") == "table" and "