Agentic RAG Chatbot

A production-style, locally-runnable Retrieval-Augmented Generation (RAG) chatbot with a full agentic workflow using LangGraph, ChromaDB, SQLite, and a FastAPI backend.

Architecture Overview

User ──► FastAPI /chat
              │
              ▼
       Input Guardrails
       (PII redact, injection detect)
              │
              ▼
        LangGraph Agent  ◄──── Conversation Memory (session-scoped)
         (ReAct loop)
        /      |      \
       ▼       ▼       ▼
  Vector    SQL DB   Clarify
  Search    Query    Tool
  (Chroma)  (SQLite)
        \      |      /
         ▼     ▼     ▼
       Output Synthesizer
       (with citations)
              │
              ▼
       Output Guardrails
       (confidence check)
              │
              ▼
           Response

Live Demo

Try it now: https://naresh-ragchat.onrender.com

No setup needed — just enter your OpenAI API key in the UI and start chatting.

Quick Start

Option A — Docker (Recommended)

cp .env.example .env          # leave OPENAI_API_KEY blank (users provide via UI)
make docker-up                # builds + starts everything
make seed                     # seeds the SQLite DB with sample data
open http://localhost:8000    # chat UI

Option B — Local (virtualenv)

python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
make seed
make dev

Supported File Types

Format	Ingestion Method	Notes
PDF	pypdf (page-aware)	Preserves page numbers in citations
TXT	Plain text read	Chunked by sentence
DOCX	python-docx	Paragraph-level chunking
JSON	Key-value flattening	Each top-level key → chunk
CSV	Row-by-row text	Column headers preserved
SQL table	SQLAlchemy introspection	Schema + data indexed

API Endpoints

Endpoint	Method	Description
/chat	POST	Send a message, get agentic response
/upload	POST	Upload a file for ingestion
/ingest	POST	Download & ingest a file from URL
/scrape	POST	Scrape a web page & ingest its text
/sources	GET	List indexed sources
/sources/{filename}	DELETE	Remove a document from the index
/sources	DELETE	Clear all knowledge base data
/sessions/{id}	GET	Retrieve conversation history
/sessions/{id}	DELETE	Clear a session
/sessions	GET	List all active sessions
/health	GET	Health check

Environment Variables

See .env.example for full list. Key variables:

• OPENAI_API_KEY — Optional server-side fallback key (users provide their own via the UI)
• MOCK_MODE — true bypasses all LLM/embedding calls
• OPENAI_BASE_URL — Override for any OpenAI-compatible API (Ollama, LM Studio, etc.)
• ENABLE_PII_REDACTION — Redact PII from user input before processing

Mock Mode (No API Key Needed)

Set MOCK_MODE=true in .env. The system uses:

• Deterministic keyword-based "LLM" responses
• Random unit-norm vectors as embeddings
• Full pipeline still runs (ChromaDB, SQLite, guardrails, memory)

This lets you test the entire system without an OpenAI key.

Deployment (Render)

This project is deployed on Render for 24/7 availability:

1. Push to GitHub
2. Connect the repo to Render as a Docker web service
3. Leave OPENAI_API_KEY blank — users provide their own key via the UI
4. The render.yaml blueprint auto-configures all environment variables

See render.yaml for the full deployment configuration.

Per-User API Key

Users provide their own OpenAI API key through a modal in the UI. Keys are:

• Stored only in the browser's localStorage (never sent to any server-side storage)
• Sent per-request in the POST body to /chat
• Never logged or persisted on the server

This means no server-side API key is needed for the app to function.

Local LLM via Ollama

ollama pull llama3
# In .env:
OPENAI_BASE_URL=http://localhost:11434/v1
OPENAI_API_KEY=ollama
LLM_MODEL=llama3
EMBEDDING_MODEL=nomic-embed-text