📄️ Instructions
Write the prompt blocks that tell your agent who it is, how it talks and what rules it follows, and make blocks apply only in certain situations.
📄️ Tools
Connect your own APIs as HTTP tools so the agent can look things up and take actions during a conversation, and learn which built-in tools every agent gets.
📄️ Functions
Write small TypeScript functions your agent calls as tools to combine several APIs, calculate and reshape data, running in a locked sandbox with strict network rules.
📄️ Skills
Give your agent detailed step-by-step expertise for specific requests that it loads only when a conversation needs it, compatible with the SKILL.md format.
📄️ Knowledge
Add documents, web pages and text that your agent searches for every customer message, and learn how processing, search, re-processing and deleting work.
📄️ Variables and secrets
Reuse values like a support phone number across prompt blocks with variables, and give tools and Functions API keys safely with secrets.
📄️ Model and behavior
Choose your agent's model and fallback model, and set its memory, sessions, language handling, web search, fallback reply and retries.
📄️ Guardrails and privacy
Protect customers' personal data from the AI, and block prompt injection, off-topic questions, instruction leaks and offensive messages.
📄️ Testing
Chat with your agent's draft in the Test panel, pretend to be a different customer, channel or time, and read what each reply was built from.
📄️ Publishing and versions
How drafts and published versions work, what Publish requires, how to compare and restore versions, and what customers see when you publish.