Models. Pick per node, pay per token.
Plooi runs every AI node against one of three providers — Anthropic Claude, OpenAI and Google Gemini. Pick the model per node, see the cost before the run, and pay the provider directly via Bring your own key.
Closed Alpha. Some model slots may not be 100% live during alpha. The picker only shows providers you have actually connected under Settings → AI Providers.
Providers
| Provider | Where it shines | Image / video |
|---|---|---|
| Anthropic Claude | Long-form reasoning, structured strategy nodes, chained methods | — |
| OpenAI | Broad coverage, short generative tasks, vision input | DALL·E |
| Google Gemini | Fast text, low cost-per-token | Imagen and Veo — the only route to Plooi’s image and video nodes |
Image generation (Imagen) and video generation (Veo) are only available through Gemini. Connect a Gemini key if you want to use image or video nodes.
Model categories
| Category | Examples | Strengths | Cost profile |
|---|---|---|---|
| Text | Claude Sonnet, GPT-4o, Gemini 2.5 Flash, Gemini 3 Pro Preview | Quick drafts through to deep reasoning | Low–medium |
| Image | Gemini 2.5 Flash Image, Imagen 3.0 | Concept sketches through to high-fidelity | Medium–high |
| Video | Veo 3.1 Fast, Veo 3.1 Standard | Previews through to final clips | High |
| Audio | Gemini TTS | Voiceover and narration | Low |
The model lineup shifts as providers release new versions. The picker in the canvas always reflects what’s live today.
Picking a model per node
- Text nodes use text models
- Image nodes use image models (Gemini only)
- Video nodes use video models (Gemini only)
- Audio nodes use audio models
The per-node choice persists across re-runs — set Claude on a strategy node once and it stays Claude on the next fold.
Cost transparency
Before every run, Plooi previews the credit cost on the node. Quality versus budget — your call, every time. Keys, encryption and storage detail live in Bring your own key.
The glass box
Every output ships with its full trace pinned alongside — no hidden hallucinations. Click any output and you see the model that produced it, the confidence it claimed, the exact prompt that ran, and the source quote it folded in.
┌────────────────────────────────────────────────┐
│ Output │
├────────────────────────────────────────────────┤
│ "Audience skews mid-market B2B with a strong │
│ preference for tools that consolidate the │
│ research → strategy hand-off…" │
├────────────────────────────────────────────────┤
│ ● confidence ████████░░ 82% │
│ ● model Claude Sonnet 4.5 │
│ ● cost 12 credits │
│ ● source Interview 03 · line 47 │
│ ● prompt [ show full prompt ] │
└────────────────────────────────────────────────┘This is the part the team actually trusts. A persona claim with the interview line pinned is reviewable; a persona claim without it isn’t.