Build Hub
Build anything across every model surface. Let it run automatically, design the join strategy yourself, or open the full studio.
Build Anything
Describe any multi-modal goal — images, video, narration, interactive games. The orchestrator designs a graph across every model surface and builds all the artifacts on demand.
Join strategies
Different modalities (text + image + audio + video) are produced in parallel, then a join assembles them into one composite artifact — e.g. an illustrated, narrated page.
LLM copy + image illustration + speech narration → one narrated page.
A single surface is invoked N times in parallel (one per page/section/variant), then a join collects all takes into a gallery or document.
N image generations (one per page) → join into a picture book.
Several surfaces answer the SAME question independently, then a join reconciles them into a single verified answer.
LLM (web context) + webSearch + a domain agent → reconciled answer.
A backend function supplies hard facts/data, a model supplies narrative or visualization, joined into a grounded report.
yfinanceApi data + LLM narrative → grounded market report.
A sequential chain where each node transforms the previous output into a new modality or representation.
TranscribeAudio → LLM summary → GenerateSpeech recap.
A complex node decomposes into a child Joined Chain Graph (with its own internal joins) whose merged result bubbles up to the parent.
A 'research' node spawns a child graph of sub-investigations.