AI in Action

Core Flow Tools' shared AI Companion panel lets any application expose a small, curated, human-reviewed API instead of raw scene access — a request comes in as natural language, the model proposes Python against that curated surface, and nothing runs until a person reads it and clicks Run (see AI Companion and ClipsAPI for the full architecture). This page is a running log of real requests captured this way, across whichever application they came from — not scripted demos, and not a one-time gallery. New cases get added here as they're recorded.

AnimHub

Prompt

Insert a 100-frame AnimHub clip at frame 221 named “Navigation_Offset_Test”. Offset all subsequent animation clips by 100 frames, and if objects are selected, offset their keyframes by the same amount.

AnimHub inserting a new 100-frame clip at frame 221, every later clip's range shifted 100 frames later, and the selected object's keyframes shifted to match

One request, three coordinated edits: a new clip created at an exact frame (clips.create()), every clip starting after it pushed 100 frames later to make room (clips.list() + clips.set_range() per clip), and the selected object's own keyframes shifted by the same amount (clips.offset_keys_after(), documented as “THE way to make room for a newly-inserted clip in the middle of an existing timeline” — a dedicated curated call rather than a hand-written per-key loop, for exactly this scenario). Full reference in ClipsAPI.

AnimHub

Prompt

I’ve many animation clips in Time Editor, offset them so they don’t overlap, leaving a 2 frame gap between the clips, then sync the Time Editor to AnimHub, replacing any takes in AnimHub, and baking the keys back to the scene.

Overlapping Time Editor clips repositioned end-to-end with a 2-frame gap between each, then AnimHub's own clip list replaced from the retimed Time Editor and the result baked back onto the scene's keyframes

A request that spans two different systems in one pass. Time Editor's own clips aren’t part of ClipsAPI's curated surface at all (they're a separate native Maya data structure from AnimHub's own CF clips), so the model reads and repositions them directly against Maya's Time Editor commands to compute the non-overlapping, 2-frame-gapped layout - exactly the kind of request the safety model is built for: no curated method covers it, so it falls back to precise, reviewable native calls instead of inventing one. The second half reuses real, existing CoreTools code rather than hand-rolled logic: the same time_editor_sync.py read (list_time_editor_clips()) and bake (bake_clips_to_scene()) functions that back AnimHub's own human-facing Tools > TimeEditor > Sync Clips to Timeline Editor... dialog, driven here without the modal dialog in the way since the whole point is a script the AI can propose and a human can review in one read, not a UI it has to click through.

Batch

Prompt

Take the file in queue, if it’s an image file, use FFMPEG to compress/re-scale until the file is less than 25MB save in the same folder and Name the file with a “Version Number” check file size after generating, if bigger than 25MG do the process again. Don’t run more than 20 times.

CF Batch's Process Queue running an AI-authored ffmpeg compression script against a queued animated GIF, producing a versioned, size-checked output under the 25MB limit

Written to solve a real, immediate need while building this documentation: the GIFs used throughout this site (including the two above) start out too large to embed comfortably, and needed compressing without hand-running ffmpeg flags each time. Describing the constraint in plain English and letting CF Batch's own AI Companion turn it into a queued, reusable script turned a one-off manual chore into an automated one. The model calls batch.create_script() + batch.queue_script() to register a real library entry (not a one-shot inline script), which itself resolves ffmpeg through ffmpeg_tool.resolve_ffmpeg() rather than assuming a bare "ffmpeg" is on PATH, and - because the target here is an animated GIF specifically - uses the same two-pass palettegen/paletteuse technique documented in Batch's own AI Companion integration rather than a naive scale-only re-encode, which bands badly on a GIF's limited palette. Each attempt re-checks the resulting file size and tries again at a smaller scale, up to the requested cap, stopping the moment it's under 25MB.

More cases land here as they're captured across AnimHub, Character Setup, and every other application that binds into the AI Companion — this page grows with the suite, it isn't scoped to one app.