> ## Documentation Index > Fetch the complete documentation index at: https://urantia.dev/llms.txt > Use this file to discover all available pages before exploring further. # Building a Urantia Book AI Chatbot with RAG > Step-by-step tutorial for building an AI chatbot that answers questions about the Urantia Book using RAG (Retrieval-Augmented Generation) with the Urantia Papers API and OpenAI. This tutorial walks through building an AI chatbot that can answer questions about the Urantia Book with accurate citations, using the Urantia Papers API for retrieval and OpenAI for generation. ## Architecture The chatbot follows the RAG (Retrieval-Augmented Generation) pattern: 1. **User asks a question** about the Urantia Book 2. **Search** the Urantia Papers API for relevant passages 3. **Retrieve context** around the top results 4. **Generate** an answer using an LLM with the retrieved passages as context 5. **Return** the answer with source citations ## Prerequisites * Node.js 18+ or Python 3.10+ * An OpenAI API key (for the LLM) * No Urantia API key needed (it's free and open) ## Step 1: Search for Relevant Passages ```typescript theme={null} async function searchUrantia(query: string, limit = 5) { const response = await fetch('https://api.urantia.dev/search', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ q: query, type: 'and', limit }) }); const data = await response.json(); return data.results; } ``` The search endpoint supports three modes: * `and` — All words must appear (best for specific queries) * `or` — Any word can appear (best for broad exploration) * `phrase` — Exact phrase match (best for quoting) ## Step 2: Get Surrounding Context The context endpoint is critical for RAG quality. A single paragraph often lacks the full meaning — surrounding paragraphs provide narrative flow. ```typescript theme={null} async function getContext(ref: string, window = 3) { const response = await fetch( `https://api.urantia.dev/paragraphs/${ref}/context?window=${window}` ); return response.json(); } ``` The `window` parameter (1-10) controls how many paragraphs before and after the target are included. ## Step 3: Build the Prompt ```typescript theme={null} async function buildPrompt(question: string) { // Search for relevant passages const results = await searchUrantia(question, 5); // Get context for top 3 results const contexts = await Promise.all( results.slice(0, 3).map(r => getContext(r.standardReferenceId, 2)) ); // Format passages for the LLM const passages = contexts.map(ctx => { const paragraphs = ctx.paragraphs .map(p => `[${p.standardReferenceId}] ${p.text}`) .join('\n\n'); return paragraphs; }).join('\n\n---\n\n'); return `You are a knowledgeable assistant about the Urantia Book. Answer the user's question based ONLY on the provided passages. Always cite specific paper references (e.g., Paper 107:0.2) for your claims. If the passages don't contain enough information to answer, say so. ## Relevant Passages from the Urantia Book ${passages} ## User Question ${question}`; } ``` ## Step 4: Generate the Answer ```typescript theme={null} import OpenAI from 'openai'; const openai = new OpenAI(); async function askUrantia(question: string) { const prompt = await buildPrompt(question); const completion = await openai.chat.completions.create({ model: 'gpt-4o', messages: [ { role: 'system', content: prompt }, { role: 'user', content: question } ], temperature: 0.3, // Lower temperature for factual accuracy }); return completion.choices[0].message.content; } // Example usage const answer = await askUrantia('What are Thought Adjusters?'); console.log(answer); ``` ## Step 5: Add Streaming (Optional) For a better user experience, stream the response: ```typescript theme={null} async function askUrantiaStream(question: string) { const prompt = await buildPrompt(question); const stream = await openai.chat.completions.create({ model: 'gpt-4o', messages: [ { role: 'system', content: prompt }, { role: 'user', content: question } ], temperature: 0.3, stream: true, }); for await (const chunk of stream) { const content = chunk.choices[0]?.delta?.content || ''; process.stdout.write(content); } } ``` ## Tips for Better Results 1. **Use `type: "and"` for specific questions** and `type: "or"` for exploratory ones 2. **Increase the context window** for complex topics — `window=5` gives broader narrative context 3. **Search with key terms** from the Urantia Book's vocabulary (e.g., "Thought Adjuster" instead of "inner spirit") 4. **Filter by paper** when you know the relevant section — use the `paperId` parameter 5. **Lower the temperature** (0.2-0.4) for factual accuracy; raise it (0.6-0.8) for more creative explanations ## Full Example (Python) ```python theme={null} import requests import openai def search_urantia(query, limit=5): r = requests.post("https://api.urantia.dev/search", json={"q": query, "type": "and", "limit": limit}) return r.json()["results"] def get_context(ref, window=3): r = requests.get(f"https://api.urantia.dev/paragraphs/{ref}/context?window={window}") return r.json() def ask_urantia(question): results = search_urantia(question, 5) contexts = [get_context(r["standardReferenceId"], 2) for r in results[:3]] passages = "\n\n---\n\n".join([ "\n\n".join([f'[{p["standardReferenceId"]}] {p["text"]}' for p in ctx["paragraphs"]]) for ctx in contexts ]) response = openai.chat.completions.create( model="gpt-4o", messages=[{ "role": "system", "content": f"""Answer based ONLY on these Urantia Book passages. Cite references. {passages}""" }, { "role": "user", "content": question }], temperature=0.3 ) return response.choices[0].message.content print(ask_urantia("What happens after death according to the Urantia Book?")) ``` ## Next Steps * Add conversation history for multi-turn chat * Implement a web UI with React or Next.js * Add audio playback for cited passages using the `/audio` endpoint * Deploy as a Telegram or Discord bot See the full recommended workflow for AI agent integration.