rag-tui 0.0.4b1

Interactive terminal UI for visualizing, debugging, and tuning RAG chunking pipelines

RAG-TUI v0.0.4 Beta

The terminal sidekick for chunking, retrieval, and “why did my RAG answer do that?”

You know the moment when your LLM confidently answers with something you never said? Yeah. RAG‑TUI exists for that exact moment. It lets you see your chunks, tweak them in real time, and test retrieval before you ship.

Now with a headless CLI and a Python API — because sometimes you want the debug UI, and sometimes you want a CI‑friendly report.

---

What Is This?

RAG‑TUI is a beautiful terminal‑based debugger for Retrieval‑Augmented Generation pipelines. It helps you tune chunking strategies, inspect retrieval results, run batch tests, and export production‑ready configs.

In short: fewer hallucinations, more receipts.

---

Quick Start (30 seconds)

Install

pip install rag-tui

Launch the TUI

rag-tui

Pro tip: Press L in the app to load a sample document and start experimenting immediately.

---

New: Headless CLI

For scripts, CI, and “just give me JSON”.

Chunk text

rag-tui chunk --file docs.txt --strategy token --chunk-size 200 --overlap-percent 10 --format json

Evaluate retrieval

rag-tui eval --file docs.txt --queries-file queries.txt --top-k 3 --threshold 0.5

Export config

rag-tui export --strategy recursive --chunk-size 600 --overlap-percent 15 --format langchain

If you want the UI explicitly:

rag-tui ui

---

New: Python API (Headless Mode)

Use RAG‑TUI directly inside notebooks or pipelines.

from rag_tui import api

# Chunk text
result = api.chunk(
    text="Hello world. Here's a tiny document.",
    strategy="sentence",
    chunk_size=200,
    overlap_percent=10,
)

# Evaluate retrieval
metrics = api.eval(
    queries=["What is this about?"],
    docs="Hello world. Here's a tiny document.",
    strategy="token",
    chunk_size=200,
    overlap_percent=10,
)

# Export config
langchain_code = api.export(
    format="langchain",
    strategy="recursive",
    chunk_size=600,
    overlap_percent=15,
)

---

What You Can Do in the TUI

1) Visualize chunking strategies

Pick from six chunkers and instantly see the results:

• Token
• Sentence
• Paragraph
• Recursive
• Fixed characters
• Custom (your own Python)

2) Tune chunk size + overlap live

Slide, watch, repeat. No guessing.

3) Search and inspect retrieval

Type a query and see exactly which chunks are returned.

4) Batch test

Paste a list of queries and get metrics like hit rate and average scores.

5) Export production configs

Generate LangChain or LlamaIndex splitters without hand‑translating numbers.

---

Providers Supported

Pick your flavor:

• Ollama (local, free, private)
• OpenAI
• Groq (fast)
• Google Gemini

Set your API key env vars, or just run Ollama locally and you’re good.

---

File Types Supported

.txt, .md, .py, .js, .json, .yaml, .pdf, .csv, and more. If it’s text‑ish, RAG‑TUI probably speaks it.

---

Why People Use RAG‑TUI

• “My chunks were too big, but I couldn’t see it.”
• “Retrieval looked fine… until I inspected the top‑k results.”
• “My RAG answers were drifting, and I needed a clean baseline.”

RAG‑TUI makes chunking and retrieval visible.

---

Common Workflows

Debug a RAG failure in 5 minutes

1. Load your document.
2. Try a few queries in Search.
3. Adjust chunk size + overlap.
4. Repeat until retrieval actually makes sense.

Generate a production config

1. Tune your chunks in the TUI.
2. Export LangChain or LlamaIndex config.
3. Paste into production. Done.

---

Tips and Tricks

• Press L to load sample text instantly.
• Use smaller chunks for QA, larger ones for summarization.
• If in doubt, start with the Q&A Retrieval preset.

---

Installation Notes

Python 3.10+ required.

pip install rag-tui

---

Contributing

Got ideas? We want them. Open an issue or send a PR. If you fix a weird chunking bug, we’ll probably name a preset after you.

---

License

MIT. Do cool things.