Workflows
Three ways the agent builds for you.
DataDriver’s agent doesn’t just fill in a template — it plans, shows you the plan, and waits for a yes before it spends anything or touches your comp. Here’s what that looks like for the three places you’ll meet it: building a comp from scratch, saving a layout as a reusable component, and narrating a batch. Every skill behind these lives in the skill catalog; what to install first is on the setup guide.
Build a comp
The Build Wizard turns a one-line subject into a finished, data-driven comp — type an idea or tap a starter card, and the agent seeds real or synthetic rows, resolves a matching archetype, and drafts a full plan before it builds anything. Nothing renders until you’ve reviewed that plan and confirmed it.
The flow
- Describe it. Type a one-line subject (or tap a starter like “Top-10 leaderboard”), then set row count, format, and data tier — Grounded (real, scraped) or Synthetic. Advanced options let you pin the archetype, vibe, motion, and image handling instead of leaving them on auto.
- Click Build. The agent seeds the data, resolves an archetype, and assembles a plan: comp name, slot mapping, sample rows, and — if AI images are on — a credits-per-image cost block.
- Review the plan card. Check the slot mapping and sample rows. If the image cost clears your spend ceiling, a second, explicit “Confirm N credits for M images” button appears — nothing spends on the first click alone.
- Confirm (or Adjust). Adjust returns to the form without discarding the run, so tweaking a field doesn’t mean starting over. Confirming starts the build; a live step tracker and the agent’s own commentary show progress phase by phase.
- Watch it land. The finished comp opens with its pass/fail facts shown, and the build is saved as a record you can Rebuild (same data) or Refresh (re-scrape sources into a new dated edition) later.
What it needs
- Claude Code — runs the wizard end to end.
- Higgsfield — only if you turn on AI-generated images.
- Firecrawl — only for the “Grounded (real)” data tier.
- DataDriver’s Agent tab additionally requires Valet — macOS only, for now.
Try it
top 10 NBA scorers of all time, ranked, with real photosComponents
An archetype is DataDriver’s name for a reusable snapshot of a comp’s row layout — its rules, layer visibility, and any motion or duration — captured once and reapplied anywhere. Batch automatically pulls from your saved archetypes to vary row layouts across a render, and any project archetype can be promoted into a shared library every project can use.
The flow
- Build the layout. Set up the rules, visibility, motion, and duration you want on a comp using the Layout tab.
- Name it and capture it. At the foot of the Comp layout drawer, type a name and click “Save as archetype” — or ask the agent to capture it. Either path runs the same write path; the agent’s runs as a dry-run preview first, showing an approval card with the rule count before anything is written.
- Approve the write. DataDriver snapshots the rules, visibility, and any motion/duration markers into your project’s archetype library. A name collision offers Overwrite (replacing a project entry) or Shadow (covering a starter for this project only).
- Reuse it. Click Edit on any saved archetype to check it out onto another comp — Save writes your changes back and restores the comp to base, Discard reverts. Batch resolves archetypes automatically per row when you want layout variety across a run, with no manual step per row.
- Share it. Click Promote on a project archetype to copy it into the shared Starters tier — available to every project from then on.
What it needs
- Claude Code — only for the agent-driven capture path (preview + approval card); the panel’s own “Save as archetype” button works without it.
- DataDriver’s Agent tab additionally requires Valet — macOS only, for now.
Try it
Capture this comp’s current layout as an archetype named "Lower Third — Speaker".Narration
Narration turns a Script column in your loaded data into per-row voiceover: DataDriver resolves each row’s voice, generates (or reuses cached) audio through Higgsfield text-to-speech, and applies it to the comp — extending duration, adding an audio layer, ducking other tracks, and optionally burning in captions.
The flow
- Load a script. Once your data has a Script column, the Narration section appears in the Batch tab automatically.
- Set the voice. Pick a named voice (or paste your own custom/cloned voice id), plus head/tail padding, how much to duck other audio, and whether to burn in captions.
- Preview — and see the spend first. Click Preview on the active row. If that line isn’t already cached, DataDriver asks “Generate 1 narration?” before spending anything. Generation happens once, through your own Higgsfield credits; every later reuse of that exact line is a free cache hit.
- Approve, when the agent drives it. The agent’s equivalent step always previews first — a “would generate” or “cache hit” card per row — before it ever calls the TTS provider, so nothing spends without you seeing it coming.
- Run the batch. Applying narration only ever uses audio that’s already been generated — it bakes it into each comp (duration, audio layer, ducking, captions) without spending anything further.
What it needs
- Higgsfield — narration credits bill to your own Higgsfield account, the same as AI images, never routed through DataDriver.
- Claude Code — for the agent-driven parts of the flow.
- DataDriver’s Agent tab additionally requires Valet — macOS only, for now.
Try it
Narrate every row’s Script column with the "Arthur" voice and burn in captions.