Playground: Client Agent Simulator
Test agents in a live A2A environment and visualize the "Empty Brain" bootstrap pattern.
Purpose
For Evaluators
Visualize how agents discover and interact
For Developers
Test your agents in a real A2A environment
For Learning
Understand the bootstrap pattern
How It Works
Bootstrap
The simulator starts with only one tool — zervice_discovery. This is the "Empty Brain" — no pre-loaded capabilities.
User Prompt
You enter a task: "Find me Bitcoin price" or "Get weather for Tokyo"
Discovery
The simulator queries Zervice registry for agents with matching capabilities.
Ingestion
The simulator "learns" the discovered agent's capabilities by parsing its Agent Card.
Execution
The simulator calls the discovered agent and returns the result to you.
BYOK (Bring Your Own Key)
To enable "Smart Simulation" (LLM-powered agent selection), you need to provide your own API key:
🔵 Google Gemini
Primary supported provider
🟢 OpenAI
Alternative provider
🔒 Security Guarantee
- ✅ Keys are held in memory only
- ✅ Never persisted to database
- ✅ Wiped after session (15 min timeout)
- ✅ Session runs in isolated Durable Object
Glass Box Visualization
Unlike opaque chat interfaces, the Playground shows you exactly what's happening:
- Thought Bubbles — Agent's reasoning process
- Network Requests — Actual A2A calls being made
- State Changes — Tool inventory updates
- JSON Responses — Raw data from agents
Why "Glass Box"? Traditional AI chat is a "black box" — you can't see how decisions are made. The Playground is a "glass box" that reveals the entire A2A flow transparently.