◢ Chapter 06 · Live Demo
Run the live multi-agent system.
The same architecture you read about in Chapter 5, running right here in your browser. Watch the audit log on the right; tweak the guideline before research; iterate.
◢ Teaching simulation
The research agent uses model-grounded synthesis (no live web tools wired in this environment). Production-grade web search/scraping/transcript retrieval lives in the sister project "Agentic AI Creator". The writer's evaluator-optimizer loop is fully real.
01 · Topic→
02 · Guideline→
03 · Research→
04 · Writer→
★ Done
Settings
01 · Topic
02 · Guideline
03 · Research
04 · Writer (Evaluator-Optimizer)
Drafts and reviews will stream here, version by version.
Evaluation Inspector
Once the writer pipeline runs, every reviewer pass and rewriter edit will be dissected here — scored by category, with the exact word-level changes highlighted.
◢ What you just saw
Three real ideas, end to end.
Structured outputs
The guideline and the reviews come back as JSON via tool-calling — no fragile parsing.
SSE streaming
Every token, every status update, every tool call streamed to the audit log as it happens.
Evaluator-Optimizer
The writer doesn't just draft — it grades itself and revises until clean (or budget hits).