---

name: quivr description: "Use Quivr to ingest documents into a brain, inspect storage and parser behavior, and ask or stream answers with RAG." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Quivr Repo Skill

Use this skill for the quivr-core package when a task involves building a brain from text or documents, asking questions of that brain, streaming answers, inspecting chat history, or diagnosing configuration, parser, or storage issues.

Route by task

• Build or repair ingestion, chunking, storage, save/load, or file-processing workflows -> sub-skills/brain-ingestion/SKILL.md
• Ask questions, stream answers, search a brain, inspect chat history, or tune retrieval/web search -> sub-skills/brain-qa/SKILL.md
• Need API signatures or public class/method facts -> references/api-reference.md
• Need config, environment-variable, or compatibility details -> references/configuration.md and references/compatibility.md
• Need concrete failure modes and fixes -> references/troubleshooting.md
• Need distilled examples or workflow maps -> references/examples.md and references/workflows.md

Safe start

Run the bundled checks from this skill directory:

python scripts/inspect_api.py --check-import
python scripts/text_brain_smoke.py --phase ingestion
python scripts/text_brain_smoke.py --phase qa

Current version notes

• In quivr-core 0.0.33, Brain.from_files and Brain.afrom_files are not reliable on non-empty inputs because process_files() extends a ProcessedDocument instead of its .chunks.
• Prefer SimpleTxtProcessor.process_file(...).chunks plus Brain.afrom_langchain_documents until that ingestion path is fixed.
• Brain.ask is a synchronous wrapper that still requires a run_id; use Brain.aask or Brain.ask_streaming from async code.
• Brain.save only serializes FAISS plus OpenAI embeddings.

When in doubt

Start with the narrowest sub-skill, then open the shared references if you need exact signatures, config fields, or troubleshooting detail. Keep optional parser stacks such as unstructured, Tika, and MegaParse as separate concerns unless the user explicitly asks for them.