Feature
AI Steps
LLMs as a first-class workflow primitive.
AI steps treat language models like any other action: typed inputs, typed outputs, retries, observability. Use them to score leads, classify tickets, summarize conversations, generate content, or run multi-step agents.
Structured outputs
Define the JSON schema you want back. We handle the prompt engineering, validation, and retry-on-malformed-output.
Bring your own keys
Plug in your own OpenAI, Anthropic, or Azure OpenAI keys. AI steps using your own keys do not count against your plan quota.
Multi-step agents
Compose AI steps with tool calls, memory, and branching to build agents that book meetings, draft emails, or triage support tickets.