---

name: retrieval-doc-chat description: "Document ingestion, parsing, chunking, embeddings, vector stores, retrieval, citations, URL crawling, and Lance RAG workflows." disable-model-invocation: true metadata: disco-role: operating license: MIT

Retrieval Doc Chat

Use this sub-skill for document-centered retrieval work in Langroid. It covers loading files and URLs, parsing and chunking, retrieval tuning, vector-store selection, citations, and Lance RAG query planning.

Use when

• answering questions from PDFs, DOCX, Markdown, text, spreadsheets, slides, URLs, or raw bytes
• configuring DocChatAgentConfig or DocChatAgent
• choosing parsing libraries, chunking settings, or embeddings
• comparing vector-store backends for retrieval or persistence
• debugging empty retrieval, parser extras, cleanup, or full_eval risk
• setting up Lance RAG query planning with filters and dataframe calculations

Covered surface

• DocChatAgentConfig, DocChatAgent
• parsing configs, loaders, and chunking
• embeddings and vector-store configs
• BM25, fuzzy search, reranking, and enrichment
• citations and file attachments
• URL crawling and retrieval-only flows
• LanceDocChatAgent and LanceRAGTaskCreator

Boundaries

• Provider and model configuration: ../llm-provider-config/SKILL.md
• Generic agent, task, and tool mechanics: ../agents-tasks-tools/SKILL.md
• SQL, table, and graph workflows: ../data-sql-graph-agents/SKILL.md

Runtime entry points

• references/api-reference.md
• references/doc-chat-workflows.md
• references/parsing-and-data-formats.md
• references/vector-stores-and-embeddings.md
• references/troubleshooting.md
• scripts/rag_config_smoke.py