Spaces:
Running
Running
Commit History
Yesterday
-
feat(documents): add bulk DELETE /documents endpoint and active task cancellation Running
9f0929b -
fix(auth): fallback to default client ID gracefully and expose shared documents
bbcb666 -
fix(postgres): safely defer owner_id index creation to auto-migration step
0deb3a0 -
fix(rag): prioritize document tree root title and exclude generic headings in research queries
1fa8476 -
feat(research): enrich academic queries with active document topic and enforce verified arXiv citations
29a9e61 -
fix(postgres): normalize DSN credentials and URL-encode special characters in password
698fc89 -
chore(deps): declare missing runtime dependencies and refresh requirements lock
841b61d -
feat(ratelimit): implement dual-key in-memory sliding window rate limiter
efdcc61 -
feat(auth): enforce client ownership and data isolation across all endpoints and databases
a506cc3 -
fix(startup): ensure static and data directories exist before mounting and writing
AI Assistant92742e1 -
feat(rag): implement nucleus top_p retrieval and exact inner page citation pinning
AI Assistant- Add temperature-scaled Softmax over weighted RRF consensus scores with top_p=0.75, min_p=0.18, and multi-hop min_k=2 - Implement exact inner-page citation pinning (active_page) during Sub-Node Paragraph Pruning from --- [PAGE X] --- markers - Update LLMService section prompt header formatting to cite active_page - Rebuild IEEE index cache with exact per-paragraph page bounds - Re-run full 95-question benchmark pushing Context Precision to 82.5% and Citation Page Accuracy to 83.1% on IEEE (+ 81.1% on FinanceBench)de0d0d6 -
feat(eval): add unified FinanceBench benchmark suite and full-system verification report
AI Assistant8b4bcb4 -
feat: upgrade FastMCP and internal tools to Research & Study MCP suite
AI Assistant- Add Deep Web Reader tool (read_webpage_markdown) with clean HTML parsing and 6s timeout - Add Academic & arXiv Research tool (search_arxiv_papers) querying public arXiv API with direct PDF links - Add YouTube Lecture Transcript tool (fetch_youtube_transcript) with 3-minute timestamped chapters and citations - Add Safe AST Math & Table Calculator (evaluate_math_expression) without eval() - Expose Atlas internal capabilities as FastMCP tools (atlas_list_documents, atlas_query_document, atlas_get_knowledge_dossier) - Expose REST router /tools with companion endpoints - Integrate arXiv research auto-triggering into VectorlessRAG and conversational stream - Update pyproject.toml, requirements.txt, and uv.lock with beautifulsoup4 and youtube-transcript-apidad455c -
feat(retrieval): implement Phase 3 query intent classification and dossier root context injection
AI Assistantf4d3bb7 -
feat(retrieval): implement vectorless hybrid search with zero-dependency BM25 and weighted RRF
AI Assistant3a2ce42 -
feat(rag): inject page boundaries and implement sub-node paragraph pruning for precise citations
AI Assistant2bc0095
11 days ago
-
feat: add FastMCP web search server, tool and dual citations
AI Assistant63f56d4 -
feat: expose database connection diagnostics on /health
AI Assistant190fb26 -
feat: integrate PostgreSQL persistence for documents, trees, and chat sessions
AI Assistant8eaad14
13 days ago
-
fix: ensure chat existence in persistent DB, safe history endpoint, and data mount support
AI Assistanta57411f -
feat: add multi-document vectorless RAG, session isolation, and memory services
AI Assistant0039ed6 -
feat: add Atlas identity, document endpoints, and Groq fallback for depleted tokens
AI Assistantdff6fcd
23 days ago
-
feat: asynchronous PDF upload and background indexing with status polling
AI Assistantd13ebb9 -
update
AI Assistantbe59b3a -
feat: add RAG application modules, services, routers, and schemas for document and chat processing
AI Assistant7b5352e
-
Refactor code structure for improved readability and maintainability
AI Assistantbdf20e7