Agent A
Agent B
Agent C
Status
API keys configured. 5 sessions running, 1 idle box.
Session status
Box status
| Prompt | Group | Model | Status | |
|---|---|---|---|---|
| Sample prompt 1 | — | Model A | completed | |
| Sample prompt 2 | Batch A | Model B | running | |
| Sample prompt 3 | Batch A | Model C | queued | |
| Sample prompt 4 | — | Model A | failed |
Create session
Mode
Single
Multi-model
Multi-prompt
Your batches
| Name | Mode | Sessions | Status |
|---|---|---|---|
| Batch A | Multi-model | 3 | 2 completed · 1 running |
| Batch B | Multi-model | 4 | 4 completed |
| Batch C | Multi-prompt | 5 | 3 completed · 2 queued |
Saved prompts
Prompt 1
Prompt 2
Prompt 3
- Environment ready
- Agent started · Model A
- Cloning starter template
- Preview ready
VaryPane
Parallel UI-generation agents in the cloud
What
Parallel UI-generation agents in the cloud. Pick a starter template, keep a reusable prompt library, attach skills, and compare models side by side — results stream in real time. Runs pi-coding-agent inside every sandbox.
Technicals
BYOK for OpenRouter and Upstash Box. Parallel agent sandboxes per session, a dedicated preview gateway, labelled box ownership, and self-hosted secret storage via Infisical.
Why
Visual comparison — and picking the best bits out of generated UIs — is usually very helpful, especially for developers doing backend work, people new to design, or anyone who just doesn't have the time for a skill this demanding. The cloud version scales and improves that workflow for hosting and for end users.
Where
Upcoming
- Screenshot-to-agent input — iterate from a wireframe or your current design
- GitHub repo connections and agent-loop iterations on one-shot designs
- Scripted environment setup