Skip to content

Troubleshooting

Anjana Tiha edited this page Mar 28, 2026 · 1 revision

Troubleshooting

Common Problems

Ollama not responding

Symptom: ConnectionRefusedError or Error: connect ECONNREFUSED 127.0.0.1:11434

Fix: Start Ollama in a separate terminal:

ollama serve

Verify it is running:

curl http://localhost:11434

ModuleNotFoundError

Symptom: ModuleNotFoundError: No module named 'chromadb' (or any other package)

Fix: The virtual environment is not activated, or dependencies are not installed:

source rag_env_311/bin/activate
pip install -r requirements.txt

Context length error / ChromaDB dimension mismatch

Symptom: InvalidArgumentError: Collection expecting embedding with dimension of X, got Y

Fix: The ChromaDB collection was built with a different embedding model. Delete it and rebuild:

rm -rf ./chroma_db/
streamlit run app.py

No chunks found

Symptom: "No supported documents found" on startup, or "I could not find relevant information" for every query.

Fix:

  • Verify files are under ./docs/ with a supported extension (.pdf, .txt, .docx, .xlsx, .xls, .pptx, .csv, .md, .html)
  • Check the terminal output for [ERROR] lines during loading
  • If using the web UI, confirm the chunk count in the sidebar is greater than 0

Model is slow

Symptom: Queries take 30–60+ seconds to respond.

Fix options:

  1. Close other applications to free RAM
  2. Use the live demo on Hugging Face (runs on cloud GPU)
  3. Reduce TOP_RETRIEVE and TOP_RERANK in src/rag/config.py (fewer chunks = faster reranking)

URL ingestion returns no chunks

Symptom: "Added 0 chunks" after pasting a URL.

Fix:

  • Verify the URL is publicly accessible (no login required)
  • Check that the page is not behind a JavaScript-rendered SPA — the system fetches raw HTML only
  • For PDFs, ensure the URL ends in .pdf or the server returns Content-Type: application/pdf

Recursive crawl indexes unrelated pages

Symptom: Pages from external domains appear in the index after a recursive crawl.

Fix: This should not happen — the crawler is restricted to the same domain as the seed URL. If it does occur, open an issue with the seed URL.


Streamlit shows a blank page or crashes on startup

Symptom: The browser shows a blank page or a Streamlit traceback.

Fix:

# Kill any existing Streamlit process
pkill -f streamlit

# Clear the Streamlit cache
rm -rf ~/.streamlit/

# Restart
streamlit run app.py

Getting Help

If none of the above resolves your issue, open an issue on GitHub with:

  1. The exact error message (copy the full traceback)
  2. Your operating system and Python version (python --version)
  3. The output of pip list | grep -E 'chromadb|ollama|streamlit'

Open an issue →

Clone this wiki locally