@fugood/bricks-project 2.25.0-beta.4 → 2.25.0-beta.40

package/types/generators/{LlmGgml.ts → LlmGgml.d.ts}

@@ -11,6 +11,8 @@
  *    - Supported Hexagon NPU for Qualcomm Snapdragon 8 Gen 1+ GPUs.
  *  - Linux / Windows  [@nextline    - Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property
  *    - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux)
+ *  - Web: Supported with CPU and WebGPU acceleration.
+ *    - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable.
  */
 import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
 import type { Data, DataLink } from '../data'
@@ -393,6 +395,31 @@ export type GeneratorLLMActionCompletion = ActionWithParams & {
         value?: boolean | DataLink | EventProperty
         mapping?: string
       }
+    | {
+        input: 'mtpSpeculativeDecoding'
+        value?: boolean | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'mtpDraftTokens'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'mtpDraftMinTokens'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'mtpDraftMinProbability'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
+    | {
+        input: 'mtpDraftSplitProbability'
+        value?: number | DataLink | EventProperty
+        mapping?: string
+      }
     | {
         input: 'functionCallEnabled'
         value?: boolean | DataLink | EventProperty
@@ -465,6 +492,8 @@ Default property:
   "cacheVType": "f16",
   "ctxShift": true,
   "cpuMoeLayers": 0,
+  "mtpSpeculativeDecoding": false,
+  "mtpDraftTokens": 3,
   "transformScriptEnabled": false,
   "transformScriptCode": "\/\* Global variable: inputs = { prompt, messages, variables }, members = { llmUtils } \*\/\nreturn inputs.prompt",
   "transformScriptVariables": {},
@@ -575,6 +604,16 @@ Default property:
     ctxShift?: boolean | DataLink
     /* Number of layers to keep MoE weights on CPU */
     cpuMoeLayers?: number | DataLink
+    /* Enable MTP speculative decoding for GGUF models that contain MTP / NextN layers */
+    mtpSpeculativeDecoding?: boolean | DataLink
+    /* Maximum number of MTP draft tokens */
+    mtpDraftTokens?: number | DataLink
+    /* Minimum number of MTP draft tokens */
+    mtpDraftMinTokens?: number | DataLink
+    /* Minimum probability for greedy MTP draft tokens */
+    mtpDraftMinProbability?: number | DataLink
+    /* Split probability for MTP speculative decoding */
+    mtpDraftSplitProbability?: number | DataLink
     /* Enable Transform Script for processing the prompt */
     transformScriptEnabled?: boolean | DataLink
     /* Code of Transform Script */
@@ -713,6 +752,7 @@ Default property:
       | {
           enabled?: boolean | DataLink
           url?: string | DataLink
+          autoDiscoverType?: 'manual' | 'auto' | DataLink
           fallbackType?: 'use-local' | 'no-op' | DataLink
           strategy?: 'prefer-local' | 'prefer-buttress' | 'prefer-best' | DataLink
         }
@@ -825,7 +865,9 @@ Default property:
    - Supported GPU acceleration by OpenCL, currently only for Qualcomm Adreno 700+ GPUs, other GPUs are not supported.  
    - Supported Hexagon NPU for Qualcomm Snapdragon 8 Gen 1+ GPUs.  
  - Linux / Windows  [@nextline    - Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property  
-   - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux) */
+   - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux)  
+ - Web: Supported with CPU and WebGPU acceleration.  
+   - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable. */
 export type GeneratorLLM = Generator &
   GeneratorLLMDef & {
     templateKey: 'GENERATOR_LLM'

package/types/generators/{LlmMlx.ts → LlmMlx.d.ts}

@@ -158,6 +158,7 @@ Default property:
       | {
           enabled?: boolean | DataLink
           url?: string | DataLink
+          autoDiscoverType?: 'manual' | 'auto' | DataLink
           fallbackType?: 'use-local' | 'no-op' | DataLink
           strategy?: 'prefer-local' | 'prefer-buttress' | 'prefer-best' | DataLink
         }

package/types/generators/{RerankerGgml.ts → RerankerGgml.d.ts}

@@ -12,6 +12,8 @@
  *  - Linux / Windows
  *    - Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property
  *    - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux)
+ *  - Web: Supported with CPU and WebGPU acceleration.
+ *    - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable.
  */
 import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
 import type { Data, DataLink } from '../data'
@@ -163,7 +165,9 @@ Default property:
    - Supported Hexagon NPU for Qualcomm Snapdragon 8 Gen 2+ GPUs.  
  - Linux / Windows  
    - Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property  
-   - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux) */
+   - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux)  
+ - Web: Supported with CPU and WebGPU acceleration.  
+   - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable. */
 export type GeneratorReranker = Generator &
   GeneratorRerankerDef & {
     templateKey: 'GENERATOR_RERANKER'

package/types/generators/{SoundRecorder.ts → SoundRecorder.d.ts}

@@ -54,7 +54,8 @@ Default property:
   "autoCuttingVocalFreqMin": 75,
   "autoCuttingVocalFreqMax": 900,
   "autoCuttingFreqClarity": 0.5,
-  "autoCuttingSpeakSpacing": 0.5
+  "autoCuttingSpeakSpacing": 0.5,
+  "userMediaAudioOptions": {}
 }
   */
   property?: {
@@ -86,6 +87,14 @@ Default property:
     autoCuttingVolumeThreshold?: number | DataLink
     /* Speak spacing of cutting fragments */
     autoCuttingSpeakSpacing?: number | DataLink
+    /* Custom getUserMedia audio options for Web / Desktop. Merged into the audio constraints. */
+    userMediaAudioOptions?:
+      | DataLink
+      | {
+          autoGainControl?: boolean | DataLink
+          echoCancellation?: boolean | DataLink
+          noiseSuppression?: boolean | DataLink
+        }
   }
   events?: {
     /* Event for chunk data (Base64 encoded PCM) */

package/types/generators/{SpeechToTextGgml.ts → SpeechToTextGgml.d.ts}

@@ -7,6 +7,8 @@
  *  - macOS: Supported GPU acceleration, recommended use M1+ chip device
  *  - Android: Currently not supported GPU acceleration (Coming soon), recommended use Android 13+ system
  *  - Linux / Windows: Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property
+ *  - Web: Supported with WASM CPU and optional WebGPU acceleration in Web Preview.
+ *    - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable.
  */
 import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
 import type { Data, DataLink } from '../data'
@@ -323,6 +325,7 @@ Default property:
       | {
           enabled?: boolean | DataLink
           url?: string | DataLink
+          autoDiscoverType?: 'manual' | 'auto' | DataLink
           fallbackType?: 'use-local' | 'no-op' | DataLink
           strategy?: 'prefer-local' | 'prefer-buttress' | 'prefer-best' | DataLink
         }
@@ -388,7 +391,9 @@ Default property:
  - iOS: Supported GPU acceleration, recommended use M1+ / A17+ chip device  
  - macOS: Supported GPU acceleration, recommended use M1+ chip device  
  - Android: Currently not supported GPU acceleration (Coming soon), recommended use Android 13+ system  
- - Linux / Windows: Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property */
+ - Linux / Windows: Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property  
+ - Web: Supported with WASM CPU and optional WebGPU acceleration in Web Preview.  
+   - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable. */
 export type GeneratorSpeechInference = Generator &
   GeneratorSpeechInferenceDef & {
     templateKey: 'GENERATOR_SPEECH_INFERENCE'

package/types/generators/{SttAppleBuiltin.ts → SttAppleBuiltin.d.ts}

@@ -1,13 +1,13 @@
 /* Auto generated by build script
  *
- * Speech recognition on iOS 26+ (Not supported on tvOS)
+ * Speech recognition on iOS 26+ and Desktop (macOS 26+). Not supported on tvOS
  *
  *  ## Features
  *  - Native Apple speech recognition
  *  - High accuracy transcription
  *  - Privacy-focused on-device processing
  *  - Supports file and data transcription
- *  - Requires iOS 26+ (Not supported on tvOS)
+ *  - Requires iOS 26+ or Desktop on macOS 26+. Not supported on tvOS
  */
 import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
 import type { Data, DataLink } from '../data'
@@ -41,6 +41,11 @@ export type GeneratorAppleSTTActionTranscribeFile = ActionWithParams & {
         value?: string | DataLink | EventProperty
         mapping?: string
       }
+    | {
+        input: 'contextualStrings'
+        value?: Array<any> | DataLink | EventProperty
+        mapping?: string
+      }
   >
 }
 
@@ -58,15 +63,26 @@ export type GeneratorAppleSTTActionTranscribeData = ActionWithParams & {
         value?: string | DataLink | EventProperty
         mapping?: string
       }
+    | {
+        input: 'contextualStrings'
+        value?: Array<any> | DataLink | EventProperty
+        mapping?: string
+      }
   >
 }
 
+/* Release context */
+export type GeneratorAppleSTTActionReleaseContext = Action & {
+  __actionName: 'GENERATOR_APPLE_STT_RELEASE_CONTEXT'
+}
+
 interface GeneratorAppleSTTDef {
   /*
 Default property:
 {
   "init": false,
   "language": "en-US",
+  "contextualStrings": [],
   "enablePartialResults": true
 }
   */
@@ -75,6 +91,13 @@ Default property:
     init?: boolean | DataLink
     /* Language/locale for recognition (e.g., 'en-US') */
     language?: string | DataLink
+    /* Phrases that should be recognized, even if they are not in the system vocabulary. */
+    contextualStrings?:
+      | Array<string | DataLink>
+      | DataLink
+      | Record<string, Array<string | DataLink> | DataLink>
+      | DataLink
+      | DataLink
     /* Enable partial results during transcription */
     enablePartialResults?: boolean | DataLink
     /* The file URL or path to be transcribed */
@@ -100,14 +123,14 @@ Default property:
   }
 }
 
-/* Speech recognition on iOS 26+ (Not supported on tvOS)
+/* Speech recognition on iOS 26+ and Desktop (macOS 26+). Not supported on tvOS
 
  ## Features  
  - Native Apple speech recognition  
  - High accuracy transcription  
  - Privacy-focused on-device processing  
  - Supports file and data transcription  
- - Requires iOS 26+ (Not supported on tvOS) */
+ - Requires iOS 26+ or Desktop on macOS 26+. Not supported on tvOS */
 export type GeneratorAppleSTT = Generator &
   GeneratorAppleSTTDef & {
     templateKey: 'GENERATOR_APPLE_STT'

package/types/generators/{ThermalPrinter.ts → ThermalPrinter.d.ts}

@@ -71,11 +71,11 @@ Default property:
   property?: {
     /* Connect printer on generator initialized */
     init?: boolean | DataLink
-    /* Connection driver */
-    driver?: 'escpos' | 'star' | 'tsc' | DataLink
-    /* The address of the printer */
+    /* Connection driver (`castles` driver: built-in or WiFi printer on Castles Saturn PDA) */
+    driver?: 'escpos' | 'star' | 'tsc' | 'castles' | DataLink
+    /* The address of the printer (For castles driver: `mode=builtin` (Saturn 1000) or `mode=wifi` (Saturn 7000); optional `heatLevel=1-5`, default 3) */
     connectString?: string | DataLink
-    /* The language of the printer (only for EPSON) */
+    /* The language of the printer (only for EPSON and TSC drivers) */
     lang?: 'ANK' | 'CHINESE' | 'TAIWAN' | 'KOREAN' | 'THAI' | 'SOUTHASIA' | DataLink
     /* The timeout of scanning */
     scanTimeout?: number | DataLink
@@ -99,10 +99,12 @@ Default property:
  { type: 'barcode', content: '1234567890', barcodeType: 'CODE128' }  
  { type: 'symbol', content: '1234567890', symbolType: 'QRCODE' }  
  Notes:  
- - `reverse-feed` only works on ESC/POS driver  
+ - `reverse-feed` works on ESC/POS driver and castles driver (line mode only)  
  - `width` and `height` in Star driver is magnification factor  
  - `y` position in ESC/POS is use feed lines to offset  
- - `textFont` only works on TSC driver */
+ - `textFont` only works on TSC driver  
+ - castles driver: `image` requires `height` in line mode  
+ - castles driver: `UPC_A`, `UPC_E`, `ITF` barcode types are not supported */
     payload?:
       | Array<
           | DataLink
@@ -149,7 +151,7 @@ Default property:
             }
         >
       | DataLink
-    /* Raw ASCII commands to print */
+    /* Raw ASCII commands to print (not supported on castles driver) */
     rawCommands?: string | DataLink
   }
   events?: {

package/types/generators/{VadGgml.ts → VadGgml.d.ts}

@@ -1,6 +1,11 @@
 /* Auto generated by build script
  *
- * Local Voice Activity Detection (VAD) inference based on GGML and [whisper.rn](https://github.com/mybigday/whisper.rn)
+ * Local Voice Activity Detection (VAD) inference based on GGML and [whisper.cpp](https://github.com/ggerganov/whisper.cpp)
+ *
+ *  ## Notice
+ *  - Web: Supported with WASM CPU in Web Preview.
+ *    - WebGPU is not used for browser VAD; the wasm package falls back to CPU for the VAD graph.
+ *    - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable.
  */
 import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
 import type { Data, DataLink } from '../data'
@@ -222,7 +227,12 @@ Default property:
   }
 }
 
-/* Local Voice Activity Detection (VAD) inference based on GGML and [whisper.rn](https://github.com/mybigday/whisper.rn) */
+/* Local Voice Activity Detection (VAD) inference based on GGML and [whisper.cpp](https://github.com/ggerganov/whisper.cpp)
+
+ ## Notice  
+ - Web: Supported with WASM CPU in Web Preview.  
+   - WebGPU is not used for browser VAD; the wasm package falls back to CPU for the VAD graph.  
+   - Single thread only in the current web preview (no SharedArrayBuffer due to cross-origin isolation / CORP), so multi-thread is unavailable. */
 export type GeneratorVadInference = Generator &
   GeneratorVadInferenceDef & {
     templateKey: 'GENERATOR_VAD_INFERENCE'

package/types/{subspace.ts → subspace.d.ts}

@@ -7,7 +7,7 @@ import type { DataCalculation } from './data-calc'
 export type Subspace = {
   __typename: 'Subspace'
   id: string
-  title: string
+  title?: string
   description?: string
   hideShortRef?: boolean
   unused?: boolean

package/utils/data.ts

@@ -85,8 +85,8 @@ type SystemDataName =
 type SystemDataInfo = {
   name: SystemDataName
   id: string
-  type: 'string' | 'number' | 'bool' | 'array' | 'object' | 'any'
-  title: string
+  type: 'string' | 'number' | 'bool' | 'boolean' | 'array' | 'object' | 'any'
+  title?: string
   description?: string
   schema?: object
   value?: any

package/utils/event-props.ts

@@ -146,6 +146,23 @@ export const templateEventPropsMap = {
     onToolChange: { BRICK_SKETCH_TOOL: 'string' },
     onExportImage: { BRICK_SKETCH_IMAGE_URI: 'string' },
   },
+  Scene3D: {
+    onLoad: { BRICK_SCENE_3D_OBJECT_ID: 'string', BRICK_SCENE_3D_DURATION_MS: 'number' },
+    onLoadError: { BRICK_SCENE_3D_OBJECT_ID: 'string', BRICK_SCENE_3D_ERROR: 'string' },
+    onObjectClick: {
+      BRICK_SCENE_3D_OBJECT_ID: 'string',
+      BRICK_SCENE_3D_HIT_POINT: '{ x?: number y?: number z?: number }',
+      BRICK_SCENE_3D_HIT_DISTANCE: 'number',
+    },
+    onObjectHover: { BRICK_SCENE_3D_OBJECT_ID: 'string', BRICK_SCENE_3D_HOVER_ENTERED: 'boolean' },
+    onAnimationEnd: { BRICK_SCENE_3D_OBJECT_ID: 'string', BRICK_SCENE_3D_ANIMATION_NAME: 'string' },
+    onFrame: {
+      BRICK_SCENE_3D_FRAME_TIME: 'number',
+      BRICK_SCENE_3D_FRAME_DT: 'number',
+      BRICK_SCENE_3D_FPS: 'number',
+    },
+    onScriptError: { BRICK_SCENE_3D_ERROR: 'string', BRICK_SCENE_3D_PHASE: 'string' },
+  },
   Tick: {
     ticking: { GENERATOR_TICK_COUNTDOWN: 'number', GENERATOR_TICK_VALUE: 'any' },
     completed: { GENERATOR_TICK_COUNTDOWN: 'number', GENERATOR_TICK_VALUE: 'any' },

package/utils/id.ts

@@ -1,38 +1,83 @@
 import { v4 as uuid } from 'uuid'
+import { createHash } from 'node:crypto'
+import { readFileSync } from 'node:fs'
+import { join } from 'node:path'
 
 let count = 0
 
-// Used for snapshot mode
-const countUUID = () => {
-  const current = count
-  count += 1
-  return `00000000-0000-0000-0000-${current.toString().padStart(12, '0')}`
+type IdType =
+  | 'animation'
+  | 'brick'
+  | 'canvas'
+  | 'generator'
+  | 'data'
+  | 'switch'
+  | 'property_bank_command'
+  | 'property_bank_calc'
+  | 'dynamic-brick'
+  | 'automation_map'
+  | 'test'
+  | 'test_case'
+  | 'test_var'
+  | 'subspace'
+
+type IdOptions = {
+  snapshotMode?: boolean
+}
+
+const APPLICATION_ID_FALLBACK = 'unknown-application'
+
+const usedStableAliases = new Set<string>()
+let applicationIdCache: { cwd: string; id: string } | null = null
+
+const readApplicationId = () => {
+  const cwd = process.cwd()
+  if (applicationIdCache?.cwd === cwd) return applicationIdCache.id
+
+  let id = APPLICATION_ID_FALLBACK
+  try {
+    const app = JSON.parse(readFileSync(join(cwd, 'application.json'), 'utf8'))
+    if (typeof app.id === 'string' && app.id) id = app.id
+  } catch {
+    // `makeId` is also used from tests and utilities outside project roots.
+  }
+
+  applicationIdCache = { cwd, id }
+  return id
+}
+
+const hashToRandomBytes = (parts: string[]) => {
+  const hash = createHash('sha256').update(JSON.stringify(parts)).digest()
+  return new Uint8Array(hash.subarray(0, 16))
 }
 
-// Make a random id if not want to use fixed id
-export const makeId = (
-  type:
-    | 'animation'
-    | 'brick'
-    | 'canvas'
-    | 'generator'
-    | 'data'
-    | 'switch'
-    | 'property_bank_command'
-    | 'property_bank_calc'
-    | 'dynamic-brick'
-    | 'automation_map'
-    | 'test'
-    | 'test_case'
-    | 'test_var'
-    | 'subspace',
-  opts?: {
-    snapshotMode?: boolean
-  },
-) => {
+const makeStableUuid = (type: string, alias?: string) => {
+  const applicationId = readApplicationId()
+  const seed = alias ?? String(count)
+  if (alias === undefined) count += 1
+
+  if (alias !== undefined) {
+    const aliasKey = JSON.stringify([applicationId, type, alias])
+    if (usedStableAliases.has(aliasKey)) {
+      throw new Error(`Duplicate makeId alias '${alias}' for type '${type}'`)
+    }
+    usedStableAliases.add(aliasKey)
+  }
+
+  return uuid({
+    random: hashToRandomBytes([applicationId, type, seed]),
+  })
+}
+
+// Make stable ids by default; explicit snapshotMode: false preserves the random escape hatch.
+export const makeId = (type: IdType, aliasOrOpts?: string | IdOptions, opts?: IdOptions) => {
   if (type === 'subspace') {
     throw new Error('Currently subspace is not supported for ID generation, please use a fixed ID')
   }
+
+  const alias = typeof aliasOrOpts === 'string' ? aliasOrOpts : undefined
+  const options = typeof aliasOrOpts === 'string' ? opts : (aliasOrOpts ?? opts)
+
   let prefix = ''
   switch (type) {
     case 'animation':
@@ -76,5 +121,11 @@ export const makeId = (
       break
     default:
   }
-  return `${prefix}${opts?.snapshotMode ? countUUID() : uuid()}`
+
+  const useCountFallback = aliasOrOpts === undefined && opts === undefined
+  const id =
+    alias !== undefined || options?.snapshotMode || useCountFallback
+      ? makeStableUuid(type, alias)
+      : uuid()
+  return `${prefix}${id}`
 }

package/skills/bricks-ctor/rules/buttress.md

@@ -1,156 +0,0 @@
-# Buttress (Remote Inference)
-
-Backend system for offloading compute-intensive AI generator tasks from BRICKS devices to more powerful machines.
-
-## Purpose
-
-When mobile devices or embedded systems lack hardware for local AI inference (LLM, speech-to-text), Buttress transparently delegates work to a server with appropriate resources (GPU).
-
-## How It Works
-
-1. **Capability Exchange**: Client and server share hardware capabilities
-2. **Strategy Selection**: System decides local vs. remote execution
-3. **Transparent Offloading**: Generator operates same way, execution happens remotely
-
-## Supported Generators
-
-- LLM (GGML) (LlmMlx.ts) - Local Large Language Model inference with GGML
-- LLM (MLX) (LlmGgml.ts) - Local Large Language Model inference with MLX
-- Speech-to-Text (GGML) (SpeechToTextGgml.ts) - Local Speech-to-Text inference with GGML
-
-## Client Configuration
-
-In generator properties, configure Buttress settings:
-
-| Setting | Description |
-|---------|-------------|
-| `Enabled` | Toggle Buttress offloading |
-| `URL` | Buttress server URL (e.g., `http://192.168.1.100:2080`) |
-| `Fallback Type` | Action if Buttress unavailable: `use-local` or `no-op` |
-| `Strategy` | Execution preference |
-
-### Strategy Options
-
-| Strategy | Description |
-|----------|-------------|
-| `prefer-local` | Use local if capable, fallback to Buttress |
-| `prefer-buttress` | Use Buttress if available, fallback to local |
-| `prefer-best` | Auto-select based on capability comparison |
-
-## Generator Configuration Example
-
-```typescript
-import { makeId } from 'bricks-ctor'
-
-const llmGenerator: GeneratorLLM = {
-  __typename: 'Generator',
-  templateKey: 'GENERATOR_LLM',
-  id: makeId('generator'),
-  title: 'Chat LLM',
-  description: '',
-  property: {
-    modelUrl: 'https://huggingface.co/ggml-org/gemma-3-12b-it-qat-GGUF/resolve/main/gemma-3-12b-it-qat-q4_0.gguf',
-    contextSize: 8192,
-    buttressConnectionSettings: {
-      enabled: true,
-      url: 'http://192.168.1.100:2080',
-      fallbackType: 'use-local',
-      strategy: 'prefer-best',
-    },
-  },
-  events: {},
-  switches: [],
-}
-```
-
-## Server Setup
-
-### Requirements
-- [Bun](https://bun.sh) v1.3+
-- GPU recommended for LLM/STT
-
-### Installation
-
-```bash
-bun add -g @fugood/buttress-server
-```
-
-### Start Server
-
-```bash
-bricks-buttress
-# or with config
-bricks-buttress --config ./config.toml
-```
-
-### CLI Options
-
-| Option | Description |
-|--------|-------------|
-| `-p, --port` | Port (default: 2080) |
-| `-c, --config` | TOML config file path |
-| `-v, --version` | Show version |
-| `-h, --help` | Show help |
-
-### Environment Variables
-
-| Variable | Description |
-|----------|-------------|
-| `HF_TOKEN` | Hugging Face token for model downloads |
-| `ENABLE_OPENAI_COMPAT_ENDPOINT` | Set to `1` for OpenAI-compatible API |
-
-## Server Configuration (TOML)
-
-```toml
-[server]
-port = 2080
-
-[runtime]
-cache_dir = "./.buttress-cache"
-n_threads = 6
-flash_attn_type = "on"
-cache_type_k = "q8_0"
-cache_type_v = "q8_0"
-
-# LLM Generator
-[[generators]]
-type = "ggml-llm"
-[generators.backend]
-variant_preference = ["cuda", "vulkan", "default"]
-gpu_memory_fraction = 0.95
-[generators.model]
-repo_id = "ggml-org/gemma-3-12b-it-qat-GGUF"
-download = true
-n_ctx = 8192
-
-# STT Generator
-[[generators]]
-type = "ggml-stt"
-[generators.backend]
-variant_preference = ["cuda", "vulkan", "default"]
-[generators.model]
-repo_id = "BricksDisplay/whisper-ggml"
-filename = "ggml-small-q8_0.bin"
-download = true
-use_gpu = true
-```
-
-## Use Cases
-
-### Resource-Constrained Devices
-Digital signage with basic hardware offloads LLM to powerful server.
-
-### Shared GPU Resources
-Multiple devices share single GPU server for inference.
-
-### Development Testing
-Test AI features on lightweight dev machines by connecting to beefy server.
-
-## Best Practices
-
-1. **Network reliability**: Ensure stable LAN connection to Buttress server
-2. **Fallback strategy**: Configure appropriate fallback for critical features
-3. **Server monitoring**: Monitor Buttress server resource usage
-4. **Model consistency**: Ensure client and server use compatible models
-5. **Security**: Run Buttress on private network, not public internet
-6. **Latency awareness**: Account for network latency in UX design