Step 6: Test & Deploy

Step 6: Test & Deploy

Full Pipeline Test

Start the dev server:

kernova dev

Open http://localhost:3000/chat and try the full pipeline:

Input: “Write a blog post about how AI is changing software development in 2026”

Expected behavior:

1. You see ”🔍 Starting research…”
2. Wait 5-10 seconds (Researcher is searching the web)
3. You see ”🧠 Analyzing findings…”
4. Wait 3-5 seconds (Analyzer is processing)
5. You see “✍️ Writing content…”
6. Wait 5-10 seconds (Writer is composing)
7. Final article appears — well-structured, sourced, engaging

Total time: 15-30 seconds for a complete researched article.

Troubleshooting

Problem	Cause	Fix
No web results	DuckDuckGo rate limited	Add TAVILY_API_KEY to .env.local
Very slow	Three sequential API calls	Normal — each takes 5-10s
Router shows help for everything	Keywords don’t match	Use trigger words: “write”, “research”, “analyze”
Empty article	Researcher found nothing	Try a broader topic

Deploy

git init && git add -A && git commit -m "Content pipeline - multi-agent"
git remote add origin https://github.com/YOU/content-pipeline.git
git push -u origin main

kernova login --key kn_live_xxx
kernova deploy

What You Built

✅ 4 specialized agents collaborating
✅ Router pattern for intelligent request routing
✅ Sequential pipeline (research → analyze → write)
✅ Delegation between agents
✅ Progress indicators for long-running operations
✅ Web search integration for real-time data
✅ Deployed to production

Key Takeaways

1. Split by skill — Each agent does one thing well
2. Code routing is fast — Use handlers for deterministic decisions
3. Temperature matters — 0.3 for facts, 0.8 for creative
4. Delegation is powerful — ctx.delegate() connects any agents
5. Progress updates — ctx.emit() keeps users informed

Ideas to Extend

• Add a reviewer agent that fact-checks the writer’s output
• Add a scheduler that auto-generates daily content
• Add different output formats (social thread, newsletter, video script)
• Add a feedback loop — user rates the article, agents improve
• Connect to CMS — auto-publish to your blog via a tool

---

Next tutorial: Autonomous DevOps Agent → (advanced)