Roadmap

Maturity

Level	Meaning
Stable	Tested, documented, API won't break
Alpha	Works, API may change
Experimental	Proof of concept
Planned	Not yet started

Implemented

• [x] Canonical DocumentGraph model — parser-agnostic document representation (Alpha)
• [x] PDF parsing — PyMuPDF adapter with text blocks, tables, bounding boxes (Alpha)
• [x] Excel parsing — openpyxl adapter with formulas, hidden content, named ranges (Alpha)
• [x] Legacy Excel (.xls) parsing — xlrd adapter (Alpha)
• [x] CSV parsing — stdlib adapter with dialect sniffing (Alpha)
• [x] DOCX parsing — python-docx adapter with headings, tables, images, metadata (Alpha)
• [x] PPTX parsing — python-pptx adapter with slides, text, tables (Alpha)
• [x] RTF parsing — striprtf adapter with text extraction (Alpha)
• [x] 9 agent tools — profile, structure, search, get_page, get_table, list_sheets, read_range, find_cells, inspect_formula (Alpha)
• [x] Schema-driven extraction — built-in schemas for invoice, contract, financial_model, table, register_spec, data_table (Alpha)
• [x] Custom schemas — register custom extraction schemas or pass JSON Schema dicts (Alpha)
• [x] Validation engine — required fields, types, arithmetic, citations, confidence (Alpha)
• [x] Citation provenance — every value traces to source page/cell/bbox with byte ranges (Alpha)
• [x] Chunking API — fixed-size and section-aware chunking for RAG pipelines (Alpha)
• [x] Batch directory processing — upload and process all files in a directory (Alpha)
• [x] Corpus search — search across all uploaded documents (Alpha)
• [x] Markdown export — render DocumentGraph as markdown (Alpha)
• [x] REST API — full FastAPI endpoint coverage with auth and rate limiting (Alpha)
• [x] Python SDK — synchronous ADX client (Alpha)
• [x] Async Python SDK — AsyncADX client with thread pool dispatch (Alpha)
• [x] MCP server — Model Context Protocol server with 10 tools (Alpha)
• [x] CLI — upload, profile, extract, validate, export, serve (Alpha)
• [x] File-based storage — JSON serialization of graphs and extractions (Alpha)

Up Next

• [ ] LLM-powered extraction (optional enhancement)
• [ ] HTML parser adapter
• [ ] Stable API guarantees (v1.0)

Not Yet Done

• No stable API guarantees — field names and response shapes may change before v1.0
• No LLM integration — extraction is heuristic-based for now
• No OCR — scanned PDFs require pre-processing
• No database storage — file-based only in MVP