What RAGvidence does
Diagnose and compare RAG retrieval techniques on your own data.
v0.2.1Local CLI · Offline reports
On this page
RAGvidence is a fully free Apache-2.0 CLI. Run retrieval experiments on local data before choosing what to implement in your RAG project. No Python project, RAG framework, database or manually prepared workspace is required for portable commands.
Start with your data
Terminal
ragvi bm25 ./manual.txt --query "How do I reset the device?"
ragvi compare-data ./docs --questions ./questions.txt --techniques bm25,parent-child| Inputs | What you can inspect |
|---|---|
| Sources only | Extraction, structure and chunk diagnostics; retrieval is not executed. |
| Sources and questions | Ranked passages, exact context, stage traces and resources; quality is unscored. |
| Reviewed original-span evidence | Evidence sufficiency, coverage, rank/precision metrics and paired improvements/regressions. |
What it does not prove
The CLI provides evidence for your choice. It does not modify or deploy your production RAG system. It does not judge generated answers, citations or abstention, and it does not promise a universally best technique.
The experiment lifecycle
- Install once, then run from any data folder.
- Inspect extraction and returned context.
- Compare alternatives on frozen data and questions.
- Add reviewed references for scored evaluation.
- Read failures, regressions, context size and observed timing together.
- Use dev questions for exploration, then freeze a shortlist before heldout evaluation.
Was something unclear? Send documentation feedback.