weichen and Cursor
06b9278996
docs(test): 记录候选 535 确认与全量切块验收
...
项目 6579 的正式正文已换成真实解析内容，冲击波 PDF 写入 90 条带向量的切块，小文档仍全文注入。
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 13:56:57 +00:00
weichen and Cursor
1fb59e12ef
feat(bizdev): 成功解析的材料一律写入 material_chunks
...
大小阈值只决定全文注入还是目录加检索片段，不再决定是否切块。上传后的异步解析同样跳过 60KB 门槛。
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 13:40:39 +00:00
weichen and Cursor
362b68a0a7
merge: feature/mineru-doc-parsing-2026-09-21 (agent page_range chunking)
...
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 11:09:52 +00:00
weichen and Cursor
db76a28689
feat(mineru): chunk Agent parse by page_range past 20-page cap
...
The free Agent lightweight API rejects files over 20 pages (-30003)
and the precision token is expired, so a real 28-page PDF could not
parse at all. On -30003, parse in 20-page windows and renumber chunk
anchors to global pages; report engine mode agent_chunked.
Also log skipped assets in ResolveMaterials and return a bool from
EnqueueAsync so parse_state is only set when a job is enqueued.
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 10:11:08 +00:00
weichen and Cursor
b867f4f19c
test: cover dedup hint, reparse guard, store gate and CoW contract
...
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 09:47:06 +00:00
weichen and Cursor
5f9fbb9923
fix(server): gate parse enqueue to store-backed pipeline
...
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 09:47:06 +00:00
weichen and Cursor
63f94ff23b
feat(server): expose dedup hint, parse error and reparse endpoint
...
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 09:18:20 +00:00
weichen and Cursor
18c727bb9f
feat(materialindex): parse pipeline with chunk embed and RAG search
...
EnsureParsed waits or lazily triggers MinerU, indexes oversized
markdown into material_chunks, and supports reranked retrieval.
Co-authored-by: Cursor <cursoragent@cursor.com >
2026-09-21 08:29:10 +00:00
