Publishing Checklist

Before publishing an agent to real users, review this list.

Prompt

• The agent goal is clear.
• The tone of voice is appropriate.
• The agent knows when to ask for more information.
• There are boundaries for out-of-scope topics (use the Security tab if needed).
• The prompt explains when to use each connected tool.

Model and cost

• The model chosen in the Model tab fits the flow complexity.
• Advanced parameters (history limit, response limit) match the channel.
• Credit cost was tested with real conversations.

Tools and bases

• Connected tools were tested individually.
• Credentials (Composio, HTTP, OpenRouter) are valid.
• Connected Knowledge Bases are up to date and processed (no items in error state).
• The agent does not try to call tools without enough data.

Channels (Triggers)

• At least one trigger is connected in Edit agent → Triggers.
• Public URL, Widget, Telegram, WhatsApp, or API tested with internal users.
• Welcome message (when applicable, on the Widget) is clear.
• There is a plan for cases that need a human — usually through the Human Handoff tool.

Monitoring

• Admins know where to review conversations (Conversations in the side menu).
• Analytics will be monitored during the first days.
• Tool failures will be investigated (Executions shows the log).
• There is a defined process to update prompt and knowledge base after launch.