Agents Playground Update: Turn-by-Turn Insights & Detailed Logging
The Agents Playground at agents.mediasfu.com just got a major upgrade — turn-by-turn conversation insights, detailed interaction logging, and a phone-only playground on the horizon.
What's New
🔍Turn-by-Turn Insights
Every conversation turn is now analyzed individually. See exactly what your agent heard, how it interpreted the input, what it decided, and what it responded — all in a structured timeline view.
📊Detailed Logging
Full interaction logs with timestamps, latency metrics, token usage, and decision traces. Export logs for offline analysis or integrate with your monitoring stack.
⚡Real-Time Analysis
Watch insights populate in real-time as your agent handles a conversation. No more waiting for post-call reports — debug and iterate live.
📱Phone Playground (Coming Soon)
Test your agents with real phone calls directly from the playground. Dial in, interact, and see the full analysis — just like your end users would experience.
Why This Matters
Building voice AI agents is hard. Debugging them is harder. When a conversation goes wrong, you need to understand exactly where it went wrong — not just that it did.
Before this update, the playground gave you a conversation interface and basic logs. Now it gives you a full observability layer:
User speaks — See the raw transcription, confidence scores, and any preprocessing applied Agent processes — View the prompt sent to the LLM, the context window, and the raw response Agent responds — Check TTS synthesis latency, audio quality metrics, and the final output Analysis generated — Automatic turn-level scoring for relevance, accuracy, and response timeHow to Use It
- Navigate to agents.mediasfu.com
- Configure your agent or use one of the preset templates
- Start a conversation — insights populate in the right panel automatically
- Click any conversation turn to expand the detailed analysis
- Use the Export button to download the full interaction log
Enable "Verbose Logging" in the playground settings to capture the full LLM prompt and response for each turn. This is especially useful when debugging context window issues or unexpected agent behavior.
What's Coming Next
📱 Phone-Only Playground
Test your agents with real phone calls. Dial a number, have a real conversation, and get the same full insights and analysis you see in the web playground.
📈 Analytics Dashboard
Aggregate insights across multiple conversations. Identify common failure patterns, track improvement over time, and benchmark against your targets.
Try the Updated Agents Playground
Start building and debugging voice AI agents with full turn-by-turn insights.
Open Agents Playground Read the Guide