Playground

Experiment safely, understand training states, and fine-tune model behavior before releasing updates.

Use the Playground as a safe testing area before exposing changes to users. Compare models, adjust prompts, and verify tone without touching the live agent.

• Training pending – the system is still processing new data.
• Training completed – the agent is refreshed and ready for questions.
• Failed – something went wrong; retry once the issue is resolved.

Each model balances speed, precision, and cost differently. Choose the one that matches the response quality and latency expectations for your use case.

• Speed-focused models reply quickly with concise answers.
• Accuracy-focused models spend more credits but provide richer detail.
• Budget-friendly models prioritize volume over depth.

Temperature ranges from 0–1 and dictates creativity. Lower values prioritize deterministic, factual answers; higher values invite expressive language but increase variance.

• 0.0–0.3: Reliable and safe responses.
• 0.4–0.7: Balanced tone for most assistants.
• 0.8–1.0: Exploratory or marketing-friendly replies.

Define the agent’s persona, voice, and guardrails. The system prompt keeps messaging on brand, limits scope, and makes sure the AI focuses on the right audience.

• Locks tone to match your brand voice.
• Keeps answers consistent across channels.
• Reminds the model which topics to prioritize.

AI Actions are currently in development and will roll out soon. The placeholder reminds teams that deeper integrations are on the roadmap.