FAQ quick reference
The three most common questions. Expand titles for details; full environment checklist in Prerequisites & licensing; troubleshooting in Troubleshooting.
Which license do I need? Same as screen license?
No. AI Agent workflows, trial runs, and chat-driven device control use the same stack as EC scripts and require a USB device license.
| License type | Purpose |
|---|---|
| USB device license | Scripts, AI workflows, automation tap/screenshot, etc. |
| USB screen license | Screen mirroring, mouse UI control, video injection, etc. |
Steps:
- Purchase USB device license (not screen license) from official channels or resellers
- Phone on USB → License Center → Batch license (device ID left, card number right)
- Or on User center My iOS devices → Batch bind license, then License Center → Refresh license in the control center
Also configure automation IPA or BLE/OTG per flow steps — see Prerequisites & licensing · Buy license · Sign IPA.
Must I configure an LLM? Which features need it?
Not every feature. LLM is only for “AI understands natural language and generates content”:
| Feature | LLM required? |
|---|---|
| AI chat (Chinese descriptions to operate devices) | ✅ Yes |
| AI workflow authoring (natural language generate/edit in editor) | ✅ Yes |
| Run / trial-run saved workflow | ❌ No |
| Batch run from task panel | ❌ No |
Import and run read-only .spk flow | ❌ No |
Without LLM, AI chat still accepts simple commands like list workflows, run … — see AI chat.
In toolbar "Model config" set API URL, model name, and key; pick separately for chat vs. AI authoring. See Config & data dirs · LLM config.
VLM (vision LLM) is separate from chat LLM: only VLM · visual locate and similar steps need it, in vlm_config.json — see Config · VLM.
Do workflows consume Token?
Running or trial-running saved workflows does not consume chat LLM Token, and does not require an active LLM API connection.
Scenarios that do consume Token (billed by your LLM provider):
- AI chat — model understands request, plans, calls device tools
- Workflow editor AI workflow authoring — generate or edit steps
Scenarios that don’t use chat LLM and don’t consume chat Token:
- Workflow trial run
- Task panel batch run
- Saved plaintext flows or imported
.spkflows
- OCR uses local engine (e.g. PaddleOCR NCNN), not chat Token.
- VLM steps bill separately per vision provider if configured — unrelated to chat Token.
- Image/color matching is local OpenCV — no LLM Token.
Related docs
- Prerequisites & licensing — full checklist and self-check table
- Getting started — first open and device selection
- Config & data dirs — LLM, VLM, data directory
- Troubleshooting — won’t open, won’t connect, execution failures