Logic blocks
Multi-step custom APIs — conditions, loops, webhooks, AI.
Logic blocks (custom APIs)
When entity CRUD isn't enough, a logic block is a multi-step pipeline you compose from steps — conditions, loops, webhooks, AI — exposed as one endpoint.
Step types
The building blocks of a logic block — entity operations, control flow, external calls, and AI steps.
Variables & templating
Steps pass data through a scoped variable namespace referenced with {{...}} — request input, prior step outputs, and request context.
Formulas
The formula step computes derived values with a small built-in function library — every function it supports, with real syntax.
Variable types
create_variable and set_variable work with five types — string, number, boolean, entity and collection — each with its own shape and shorthand accessors.
The connector step
How connector steps inside a logic block map fields into a pre-defined request, where auth comes from, and how to read a failure.
AI Call
ai_call is the one AI step you need — classification, extraction, summarization, generation, comparison — plus how to ground it in entity data for RAG.
AI Call — Intent
What each of the 9 intents actually changes, stated honestly — it's the system-prompt persona, not different execution logic.
AI Call — Context
The 5 context field types and exactly what happens to each at execution time — fetch, transcribe, or pass through.
AI Call — Output
The exact output data shape per format — the field names downstream steps actually see, stated precisely, plus how to use a list output as a Loop source.
AI Call — Model & config
Provider, model, max output tokens, temperature, and the extra settings a model accepts — including the models whose temperature is fixed.
AI Call — Conversation context
ai_call has no memory of its own — the two real ways to build a multi-turn chat feature on top of it, and how to pick between them.