AI helper
"AI helper" in the LmBox Bridge desktop app (sidebar label , route openaiChat) is a full chat page in the Tauri shell — not the old CEP panel inside the editing app. The chrome shows the connected host (DaVinci Resolve / Premiere / After Effects, and so on). The bottom bar switches among Chat / Agent / Images / Video, and picks Official cloud → Golden key or a Keychain model.
What you will learn
- Open AI helper from the desktop sidebar (or a Scaffolding slot)
- Switch mode, provider and image/video parameters in the bottom bar
- Read the mode · provider · model tag on each reply
- Save or import results, and understand common failure toasts
Guide scope
This is the main lab guide for the current desktop UI. The old CEP “home → Lab → AI helper” path is no longer the primary story; if you still see that wording elsewhere, prefer this page.
What it solves
| Pain | How this page helps |
|---|---|
| Need workflow, copy or planning help while editing | Chat / Agent via Official cloud or your own models |
| Need stills for mood or reference on the timeline | Images with aspect and size; save or import to the host |
| Need a short motion sample | Video with aspect, duration and resolution; player in the card |
| Want to try without your own API key | Official cloud → Golden key (sign-in; caps and entitlements apply) |
| Already have an OpenAI-compatible endpoint | Add it in Keychain, then pick it in the bottom cascader |
Before you start
- Use the LmBox Bridge desktop app (not the retired plug-in-only installer).
- Light up the host icon in the sidebar (Premiere / After Effects / DaVinci Resolve, …) so import can reach the extension.
- Official cloud: sign in first; membership and bottle caps follow on-screen rules.
- Own models: add an OpenAI-compatible profile in Keychain (chat / image / video addresses and enabled models as needed).
How to open
- Sidebar: click (
#icon-jiqiren). - Scaffolding: you can pin the same route-style capability; hosts match
openaiChat.plugIninmenu.json(Premiere CEP/UXP, After Effects CEP, DaVinci Resolve, …). - You do not need the old “extension home → Lab” entry.
Desktop UI
Chat layout: message list above; history / new chat in the top strip; mode, credentials and send in the bottom input shell.
Chat

Figure: sidebar AI helper selected; reply tag like Chat · Official cloud · Golden key; bottom bar for mode and provider.
Images

Figure: tag Images · Official cloud · Golden key; aspect/size such as 16:9, 1280×720; reference-image strip above the textarea.
Video

Figure: tag Video · 〈Keychain platform〉 · 〈model id〉 (example: another line with doubao-seedance-2.0); embedded player; save / import actions.
Regions
| Area | Role |
|---|---|
| Sidebar AI helper | Opens this page; chrome shows host context |
| History / New chat | Session drawer (pin, rename, delete, export Markdown); start fresh |
| Bubbles | User prompt + assistant reply; meta line mode · provider · model |
| Reply actions | Retry, copy, thumbs up / down (per message allow-list) |
| Image card | Save, zoom, use as reference, import to connected host |
| Video card | Player; save; import to connected host |
| Bottom shell | Text; + references in media modes; mode select; Keychain/model cascader; image aspect·long edge / video aspect·duration·resolution; send / stop |
Four modes
Bottom-left mode select (remembers last choice):
| Mode | Meaning |
|---|---|
| Chat | General Q&A; Official cloud shows Golden key as the leaf name |
| Agent | Editing-oriented agent flow (may use connected extension tools; finish HITL prompts before typing again) |
| Images | Image API path (not the chat stream); prompt required; optional local references |
| Video | Video API path; shares Keychain/model memory with the Blueprint video node |
TIP
Screenshots usually show Chat, Images and Video; Agent is in the same dropdown when you need it.
Providers and models
Cascader level one:
- Official cloud → leaf Golden key (chat, agent, image and video, subject to account capability).
- Your Keychain (name from Keychain manager) → then an enabled model id.
When creating a Keychain you can pick common presets (OpenAI, DeepSeek, Kimi, Alibaba Cloud Bailian, Volcengine, SiliconFlow, StepFun, iFlytek Spark, xAI, Mistral, Groq, Relay / Other, and more); the base URL fills in from the preset. Image / video need the matching modality configured; otherwise the UI warns that the channel is not set up.
Reply meta:
- Official:
mode · Official cloud · Golden key - Keychain:
mode · 〈platform〉 · 〈model id〉(platformcomes from that Keychain item)
Image / video parameters
Images (default 16:9, long edge 1280):
- Aspect:
1:1,16:9,9:16,4:3,3:4,3:2,2:3 - Long-edge presets:
512/768/1024/1280/1920/2048(UI shows derived width×height)
Video (default 16:9, 5 s, 720p):
- Aspect:
16:9,9:16,1:1 - Duration:
5/8/10s - Resolution:
480p/720p/1080p
In media modes, use + above the input to add local reference images (there is a cap; you can also add from a generated still via Reference).
Suggested flow
- Open AI helper; confirm the host icon is lit if you plan to import.
- Pick mode → Official cloud / Golden key or a Keychain model.
- For media modes, set aspect (and duration/resolution); add references if needed.
- Type the prompt; send with the button or Enter (Shift + Enter for a new line). Use Stop while generating.
- On the result: retry / copy / feedback; images also save, zoom, reference, import; video plays, saves, imports.
- Manage history in the drawer; export Markdown from the desktop app.
Host and import
- The menu item targets Premiere (CEP / UXP), After Effects (CEP), DaVinci Resolve, and similar; a lit icon means the desktop app is linked.
- Import uses the same lab footage path as Media Hub: into the current host. Without a connection (or outside the desktop app) you get a connect-first toast.
- Agent tool calls need the matching host online; while a HITL step is open, the input stays locked until you confirm or cancel.
Limits and failure modes
Please note
Model calls cost ongoing usage: Official cloud spends bottle caps; Keychain bills go to your provider. Availability follows your account and provider.
| Situation | Typical meaning |
|---|---|
| Official cloud without sign-in | Sign in first |
| Membership too low | Upgrade / unlock the required tier |
| No model / no API key | Pick a model in the bar, or set the Keychain API key |
| Image/video channel missing | Configure that modality on the Keychain, or switch profile |
| Image timeout | Try a smaller long edge |
| Video timeout | Try a shorter duration |
| Import failed | Connect the LmBox Bridge extension in the target app |
| Actions during streaming | Wait or click Stop |
Code blocks can still be copied; runnable script actions need the connected editing extension (follow the button hint). Do not look for the old CEP top-bar “gpt-4.0” model switch described in earlier docs.
Set up your own Keychain
- Open Keychain in desktop settings.
- Add an AI LLM profile; pick a platform preset or Relay / Other.
- Fill API Key and base URL (usually through
/v1); configure chat / image / video addresses and enabled models as needed. - Back in AI helper, pick the profile in the cascader. Each mode remembers its last Keychain + model; video shares memory with the Blueprint video node.
Video tutorial
Release notes
3.3.1Switch agent permission modes (Plan / Read-only / Standard / Trust).3.3.1Side agent can remember preferences and common settings across chats.3.3.0Open the sideAI helperfrom the footer anywhere in the app to chat with the agent while staying on the current page.2.7.0New Keychains can pick common providers such as Kimi, SiliconFlow, StepFun, iFlytek Spark, xAI, Mistral and Groq; the address fills in once you choose a platform.1.6.0Chat and agent modes can use Official cloud / Golden key without your own model key.1.6.0Image and video modes can use official Golden key generation (account capability required).0.2.5You can configure different models for chat, images, video and related uses, and pick them on this page.0.2.0Chat, Agent and Images each remember the last Keychain and model; Video shares memory with the Blueprint video node.- Docs
2026-10-22: rewritten for the Tauri desktop UI (sidebar entry, four modes, bottom-bar credentials and media params, result-card actions); CEP panel path is no longer the main guide.