Build High-Speed Agentic APIs with Qwen-3 235B and OpenKBS

Tutorial Overview

In this tutorial, you'll learn how to build high-performance agentic APIs that can process complex requests in under 3 seconds. Traditional APIs serve static responses, but agentic APIs reason through problems and construct their own workflows autonomously.

What you'll learn:
• How to deploy ultra-fast agentic APIs using Qwen-3 235B
• Real-time data processing and contextual intelligence
• Production-ready API deployment with OpenKBS
• Advanced content generation with conversation context

Prerequisites: Basic understanding of APIs and web development

Performance target: By the end of this tutorial, your APIs will execute complex multi-step workflows in 2.71 seconds ⚡

Step 1: Understanding Agentic API Performance

Your API can research any product in real-time, analyze multiple sources, generate personalized content, and return structured JSON—all while you grab coffee. Two LLM reasoning cycles, live web research, and secure orchestration happen faster than most database queries.

Here's what's now possible. This API researches products in real-time and generates personalized content:
Click the link → Hit "Deploy" → Watch the agent execute in real-time → Copy your production API endpoint directly from the OpenKBS web console
Performance comparison:
• Gemini 2.5 Pro + Google Search: 27.29 seconds
• Claude Sonnet 4 + Google Search: 22.46 seconds
• Claude 3.5 Haiku + Google Search: 10.58 seconds
• Qwen-3-235B + Google Search: 2.71 seconds ⚡ 8x faster
curl --location 'https://your-agent-id.apps.openkbs.com/api' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer YOUR_API_KEY' \
--data '{
  "message": "Search site:amazon.com for La Pavoni coffee machine. Return JSON with: {translated_products: {name, name_es, image}}. Ensure only one search call"
}'

# Copy API endpoint with keys from OpenKBS console above

Key Concepts: Agentic vs Traditional APIs

Traditional APIs:
• Serve predefined responses from databases
• Follow hardcoded logic paths
• Limited to pre-programmed scenarios

Agentic APIs:
• Research and analyze data in real-time
• Create responses specific to each request
• Maintain conversation context across calls
• Dynamically combine resources to achieve goals

The key difference: instead of pre-programming every scenario, you provide tools and describe the objective. The agent constructs its own workflow to achieve the goal.

Step 2: Building Advanced Content Generation APIs

The same agent can continue the conversation with context, generating comprehensive product content:
curl --location 'https://your-agent-id.apps.openkbs.com/api' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer YOUR_API_KEY' \
--data '{
  "message": "Search for La Pavoni EPC-8 Europiccola Lever Style Espresso Machine. Immediately after the response, Return JSON with: {translated_to_es: {name, generate_extended_article, attributes, specifications, price, links}} Ensure only one search call"
}'
The agent autonomously:

• Researches the product across multiple sources
• Generates comprehensive marketing copy in Spanish
• Extracts current pricing and validates links
• Structures everything as JSON

Step 3: Implementing Contextual Intelligence

Here's where things get interesting. Pass the chatId from your previous API call, and the agent maintains the entire conversation session. It remembers what you searched for, what data it found, and what you were trying to accomplish. Now ask for something related—recommendations, comparisons, deeper analysis—and watch it build upon everything that came before.
curl --location 'https://your-agent-id.apps.openkbs.com/api' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer YOUR_API_KEY' \
--data '{
  "message": "First think what 3 complementary products would enhance this item, then Search site:amazon.com for them in parallel. Return JSON with: {translated_products: {name, name_es, image}}.",
  "chatId": "previous-conversation-id"
}'
The agent:

• Remembers - the previous La Pavoni espresso machine context
• Reasons - about complementary products (tamper, milk frother, grinder)
• Searches - for these products in parallel
• Optimizes - the selection based on the original product
• Maintains - conversation continuity for future interactions
This is impossible with traditional APIs. No amount of pre-programming can match the contextual reasoning and adaptability of an LLM-powered agent.

Step 4: Deploy Your Production API

Live examples to test:
Click → Deploy → Get your production API