How opendatalab/mineru works
MinerU parses documents across PDF, image, office, web, and ebook formats, using OCR/layout and specialized table, formula, and vision-language models, then exposes structured results and rendered formats. The main workflow is a user submitting a file through the CLI, API, or Gradio; a parser service manages the request and stored results. The Doclib server adds persistent ingestion, scanning, search, caching, and background work. Edges are limited to relationships shown in excerpts or the source index; links into unsampled parsing stages are marked unverified.
Main entry points: parse_svc.py Parse service · api_client.py API parser client · server.py Doclib HTTP API · api_server.py Parse API server
Components 26
User interfaces
- CLI commands
mineru/cli/main.py - Gradio app
mineru/kit/gradio/app.py - Doclib HTTP API
mineru/doclib/server.py
Document library
- Parse service
mineru/doclib/services/parse_svc.py - Document database
mineru/doclib/core/db.py - File ingestion
mineru/doclib/background/ingest.py - Scan and watch
mineru/doclib/services/scan_svc.py - Full-text search
mineru/doclib/core/fts.py - Configuration
mineru/config.py - Parse server health
mineru/doclib/background/parse_server_health.py
Parsing pipeline
- API parser client
mineru/parser/api_client.py - Parser interface
mineru/parser/base.py - Parse API server
mineru/parser/api_server.py - Format detection
mineru/filetypes.py - PDF analysis
mineru/backend/analysis/pdf/pipeline.py - Other format analysis
mineru/backend/analysis
Models and inference
- Model registry
mineru/model/registry.py - Layout and OCR
mineru/model/layout - Table recognition
mineru/model/table - Formula recognition
mineru/model/mfr - Vision-language runtime
mineru/model/vlm
Results and rendering
- Document postprocessing
mineru/backend/postprocess/document.py - Result rendering
mineru/render/api.py - HTML renderer
mineru/render/html.py
Other components
- User
- Remote parse service
Connections 26
- User → CLI commands: submits request
mineru/cli/main.py - User → Gradio app: submits file
- User → Doclib HTTP API: sends request
- CLI commands → Doclib HTTP API: manages server
- Gradio app → API parser client: parses files
mineru/kit/gradio/app.py - Doclib HTTP API → Parse service: delegates parsing
mineru/doclib/server.py - Parse service → Document database: reads and writes
mineru/doclib/services/parse_svc.py - Parse service → API parser client: submits parse
mineru/doclib/services/parse_svc.py - File ingestion → Parse service: ensures ingestion
mineru/doclib/background/ingest.py - Parse service → Full-text search: indexes content
mineru/doclib/services/parse_svc.py - Doclib HTTP API → Full-text search: serves search
- Doclib HTTP API → Scan and watch: serves scans
- Configuration → Parse service: configures
mineru/doclib/services/parse_svc.py - Parse server health → API parser client: checks remote API
mineru/doclib/background/parse_server_health.py - API parser client → Parser interface: implements
mineru/parser/api_client.py - API parser client → Remote parse service: sends parse requests
mineru/parser/api_client.py - Parse API server → Format detection: detects formats
- Parse API server → PDF analysis: runs PDF analysis
- Parse API server → Other format analysis: handles formats
- PDF analysis → Vision-language runtime: uses inference
mineru/backend/analysis/pdf/pipeline.py - PDF analysis → Layout and OCR: extracts geometry
mineru/backend/analysis/pdf/pipeline.py - Model registry → Table recognition: registers models
mineru/model/registry.py - Model registry → Vision-language runtime: selects runtime
mineru/model/registry.py - API parser client → HTML renderer: renders HTML
mineru/parser/api_client.py - Document postprocessing → Result rendering: prepares results
- Result rendering → HTML renderer: renders HTML
mineru/render/html.py