Skip to content

brew-ai Overview

brew-ai is a conversational assistant docked inside the model builder. The legacy application had no AI assistant at all — brew-ai is new capability layered directly onto the canvas, so you can ask questions about the workflow you’re building, run functionals, schedule tasks, and get code help without leaving Build.

Docked brew-ai chat panel showing a streaming response with the thinking-chain tool-step timeline expanded

The chat panel docks alongside the canvas and streams responses token-by-token over SSE as the assistant replies, rather than waiting for a full response before showing anything. While the assistant is working, a “thinking chain” shows the server-side tool steps it’s running — reading the model, inspecting a functional, sampling output rows, and so on — each with a live elapsed-time counter while it’s in progress and a done/error state once it settles. Assistant replies render as GFM markdown, so tables, code blocks, and formatted lists all display properly. You write your side of the conversation in a rich-text composer (built on TipTap) that supports @-mentioning model inputs, outputs, and components directly in your message.

The chat panel is feature-flagged at the system level, and individual capabilities can be marked experimental — visible only to users who’ve opted into experimental features. If you don’t see the assistant in Build, or a capability described here isn’t showing up, check with your administrator about your deployment’s configuration. Responses are always generated by the real AI backend in normal use; a mock response mode exists only for offline frontend development and isn’t something you’ll encounter in a deployed environment.

Beyond free-form chat, brew-ai supports slash commands like /explain, /schedule, /find, and /execute-functional, plus @-mentions that reference model inputs, outputs, and components — hovering a mention highlights the matching component on the canvas. See Commands & Mentions for detail.

Ask brew-ai to run a functional and it walks you through a picker, an editable run card for scalar inputs and CSV uploads, and an async run — then narrates the results back to you once they’re in. See Running Functionals in Chat for detail.

Ask brew-ai to schedule a workflow and it proposes an editable task card — name, cron schedule with live validation, and email-on-error — before creating the task. See Scheduling Tasks via Chat for detail.

For Python, SQL, R, and JS code functionals, brew-ai can read your live code, LSP diagnostics, and typed inputs/outputs together, then propose edits as an inline diff you can apply or reject. See Code Copilot for detail.