import { createConfigSchematics } from "@lmstudio/sdk";

export const defaultPluginSettings = {
  PREVIEW_IN_CHAT: true,
  HTTP_SERVER_PORT: 54760,
  embeddingBaseUrl: "http://127.0.0.1:1234/v1",
  embeddingApiKey: "",
  qwen3VlModelPath: "qwen/qwen3-vl-8b",
  visionPrompt: "",
  embedPngMetadata: true,
  serverMaxTokens: 768,
  serverTemperature: 0.7,
  qwen3VlOdPrompt: [
    "Detect objects in the image with strict hierarchical prioritization.",
    "",
    "PRIORITY 1 (CRITICAL - MUST DETECT FIRST):",
    '- You MUST detect "human face" (highest priority if a person is present)',
    '- You MUST detect "person" (if no face is clearly visible or if the person is the main subject)',
    "",
    "PRIORITY 2 (MAIN SUBJECT / HERO ELEMENT):",
    "- The most visually prominent object or subject that is NOT part of the background.",
    "- Use specific, concrete labels (e.g., 'red car', 'fluffy owl toy').",
    "- Avoid generic terms like 'object' or 'thing'.",
    "",
    "PRIORITY 3 (CONTEXTUAL BACKGROUND ELEMENTS):",
    "- Only detect background elements if they are significant to the scene composition OR if the main subject is interacting with them.",
    "- Do not detect minor or redundant background details.",
    "",
    "PRIORITY 4 (FOCUSSED MAIN SUBJECT / HERO ELEMENT):",
    "- All visible body parts (hands, feet, arms, legs).",
    "- Elements of the face, as far as clearly detectable and focussed on close-ups: nose, mouth, left and right eyes, eyebrows and ears",
    "- anatomical details, as far as recognizable as \\\"focussed\\\" or \\\"prominent\\\" (e.g., 'iris', 'pupil', 'eyelid')",
    "",
    "RULES:",
    "- Maximum 16 objects total.",
    "- Each bounding box must be unique and non-redundant.",
    "- For clothing, name the specific garment (e.g., 'tank top', 'jeans').",
    "- For body parts, qualify by position (e.g., 'left hand').",
    "- NEVER prioritize background elements over the main subject or human face.",
    "- NEVER prioritize anatomical details over general concepts unless they are solely focussed (e.g. only detect 'eyes' unless 'human face' is the dominant part of the image)",
    "- If the main subject is a person, focus on the person and their immediate interactions. Ignore background elements unless they are directly involved in the interaction.",
    "- If NO person or face is visible, ALWAYS detect Priority 2 and Priority 3 subjects regardless.",
  ].join("\n"),
  detectMaxTokens: 2048,
  detectTemperature: 0.3,
  includeGenerationMetadata: true,
} as const;

export const globalConfigSchematics = createConfigSchematics()
  .field(
    "PREVIEW_IN_CHAT",
    "boolean",
    {
      displayName: "Previews in Chat",
      subtitle: "When enabled, tool responses include inline image previews. Recommended for local models without vision capability.",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.PREVIEW_IN_CHAT
  )
  .field(
    "HTTP_SERVER_PORT",
    "numeric",
    {
      displayName: "Local HTTP Server Port",
      subtitle: "Port for serving generated images over localhost (default: 54760).",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.HTTP_SERVER_PORT
  )
  .field(
    "embeddingBaseUrl",
    "string",
    {
      displayName: "Vision API Base URL",
      subtitle: "OpenAI-compatible /v1 URL. Vision tools use the same server root and call LM Studio's internal /api/v1 vision endpoints. Separate from the agent API.",
      placeholder: "http://127.0.0.1:1234/v1",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.embeddingBaseUrl
  )
  .field(
    "embeddingApiKey",
    "string",
    {
      displayName: "Vision API Key",
      subtitle: "Optional key for the Qwen3-VL vision backend. Separate from the agent API key.",
      isProtected: true,
      placeholder: "sk-...",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.embeddingApiKey
  )
  .field(
    "qwen3VlModelPath",
    "string",
    {
      displayName: "Qwen3-VL Model",
      subtitle: "LM Studio model key for the Qwen3-VL Vision API backend, for example qwen/qwen3-vl-8b. This is not a filesystem path.",
      placeholder: "qwen/qwen3-vl-8b",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.qwen3VlModelPath
  )
  .field(
    "visionPrompt",
    "string",
    {
      displayName: "Vision Prompt",
      subtitle: "Default prompt sent to the vision model when the agent does not supply one. Leave empty to disable automatic visual description.",
      placeholder: "Analyze this image based strictly on what is directly visible. Do not infer, assume, or complete information that is not present.",
      isParagraph: true,
    },
    defaultPluginSettings.visionPrompt
  )
  .field(
    "embedPngMetadata",
    "boolean",
    {
      displayName: "Embed Metadata in PNGs",
      subtitle: "Write analysis provenance, detected objects, and bounding boxes into saved PNGs as Draw Things-compatible XMP metadata.",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.embedPngMetadata
  )
  .field(
    "includeGenerationMetadata",
    "boolean",
    {
      displayName: "Include Generation Metadata",
      subtitle: "When enabled, Draw Things generation parameters (prompt, model, sampler, seed, ...) embedded in PNG files are appended to each analysis result.",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.includeGenerationMetadata
  )
  .field(
    "serverMaxTokens",
    "numeric",
    {
      displayName: "Vision API: Max Tokens",
      subtitle: "Maximum response length in tokens (1-4096). Default: 768.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.serverMaxTokens
  )
  .field(
    "serverTemperature",
    "numeric",
    {
      displayName: "Vision API: Temperature",
      subtitle: "Sampling temperature (0.0-2.0). Default: 0.7.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.serverTemperature
  )
  .field(
    "qwen3VlOdPrompt",
    "string",
    {
      displayName: "Qwen3-VL: Object Detection Prompt",
      subtitle: "Instruction sent to Qwen3-VL for default object detection. Leave empty to use the built-in default.",
      placeholder: "",
      isParagraph: true,
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.qwen3VlOdPrompt
  )
  .field(
    "detectMaxTokens",
    "numeric",
    {
      displayName: "Vision API Detect: Max Tokens",
      subtitle: "Maximum response length in tokens for object detection (1-4096). Default: 2048.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.detectMaxTokens
  )
  .field(
    "detectTemperature",
    "numeric",
    {
      displayName: "Vision API Detect: Temperature",
      subtitle: "Sampling temperature for object detection (0.0-2.0). Default: 0.3.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.detectTemperature
  )
  .build();

import { createConfigSchematics } from "@lmstudio/sdk";

export const defaultPluginSettings = {
  PREVIEW_IN_CHAT: true,
  HTTP_SERVER_PORT: 54760,
  embeddingBaseUrl: "http://127.0.0.1:1234/v1",
  embeddingApiKey: "",
  qwen3VlModelPath: "qwen/qwen3-vl-8b",
  visionPrompt: "",
  embedPngMetadata: true,
  serverMaxTokens: 768,
  serverTemperature: 0.7,
  qwen3VlOdPrompt: [
    "Detect objects in the image with strict hierarchical prioritization.",
    "",
    "PRIORITY 1 (CRITICAL - MUST DETECT FIRST):",
    '- You MUST detect "human face" (highest priority if a person is present)',
    '- You MUST detect "person" (if no face is clearly visible or if the person is the main subject)',
    "",
    "PRIORITY 2 (MAIN SUBJECT / HERO ELEMENT):",
    "- The most visually prominent object or subject that is NOT part of the background.",
    "- Use specific, concrete labels (e.g., 'red car', 'fluffy owl toy').",
    "- Avoid generic terms like 'object' or 'thing'.",
    "",
    "PRIORITY 3 (CONTEXTUAL BACKGROUND ELEMENTS):",
    "- Only detect background elements if they are significant to the scene composition OR if the main subject is interacting with them.",
    "- Do not detect minor or redundant background details.",
    "",
    "PRIORITY 4 (FOCUSSED MAIN SUBJECT / HERO ELEMENT):",
    "- All visible body parts (hands, feet, arms, legs).",
    "- Elements of the face, as far as clearly detectable and focussed on close-ups: nose, mouth, left and right eyes, eyebrows and ears",
    "- anatomical details, as far as recognizable as \\\"focussed\\\" or \\\"prominent\\\" (e.g., 'iris', 'pupil', 'eyelid')",
    "",
    "RULES:",
    "- Maximum 16 objects total.",
    "- Each bounding box must be unique and non-redundant.",
    "- For clothing, name the specific garment (e.g., 'tank top', 'jeans').",
    "- For body parts, qualify by position (e.g., 'left hand').",
    "- NEVER prioritize background elements over the main subject or human face.",
    "- NEVER prioritize anatomical details over general concepts unless they are solely focussed (e.g. only detect 'eyes' unless 'human face' is the dominant part of the image)",
    "- If the main subject is a person, focus on the person and their immediate interactions. Ignore background elements unless they are directly involved in the interaction.",
    "- If NO person or face is visible, ALWAYS detect Priority 2 and Priority 3 subjects regardless.",
  ].join("\n"),
  detectMaxTokens: 2048,
  detectTemperature: 0.3,
  includeGenerationMetadata: true,
} as const;

export const globalConfigSchematics = createConfigSchematics()
  .field(
    "PREVIEW_IN_CHAT",
    "boolean",
    {
      displayName: "Previews in Chat",
      subtitle: "When enabled, tool responses include inline image previews. Recommended for local models without vision capability.",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.PREVIEW_IN_CHAT
  )
  .field(
    "HTTP_SERVER_PORT",
    "numeric",
    {
      displayName: "Local HTTP Server Port",
      subtitle: "Port for serving generated images over localhost (default: 54760).",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.HTTP_SERVER_PORT
  )
  .field(
    "embeddingBaseUrl",
    "string",
    {
      displayName: "Vision API Base URL",
      subtitle: "OpenAI-compatible /v1 URL. Vision tools use the same server root and call LM Studio's internal /api/v1 vision endpoints. Separate from the agent API.",
      placeholder: "http://127.0.0.1:1234/v1",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.embeddingBaseUrl
  )
  .field(
    "embeddingApiKey",
    "string",
    {
      displayName: "Vision API Key",
      subtitle: "Optional key for the Qwen3-VL vision backend. Separate from the agent API key.",
      isProtected: true,
      placeholder: "sk-...",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.embeddingApiKey
  )
  .field(
    "qwen3VlModelPath",
    "string",
    {
      displayName: "Qwen3-VL Model",
      subtitle: "LM Studio model key for the Qwen3-VL Vision API backend, for example qwen/qwen3-vl-8b. This is not a filesystem path.",
      placeholder: "qwen/qwen3-vl-8b",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.qwen3VlModelPath
  )
  .field(
    "visionPrompt",
    "string",
    {
      displayName: "Vision Prompt",
      subtitle: "Default prompt sent to the vision model when the agent does not supply one. Leave empty to disable automatic visual description.",
      placeholder: "Analyze this image based strictly on what is directly visible. Do not infer, assume, or complete information that is not present.",
      isParagraph: true,
    },
    defaultPluginSettings.visionPrompt
  )
  .field(
    "embedPngMetadata",
    "boolean",
    {
      displayName: "Embed Metadata in PNGs",
      subtitle: "Write analysis provenance, detected objects, and bounding boxes into saved PNGs as Draw Things-compatible XMP metadata.",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.embedPngMetadata
  )
  .field(
    "includeGenerationMetadata",
    "boolean",
    {
      displayName: "Include Generation Metadata",
      subtitle: "When enabled, Draw Things generation parameters (prompt, model, sampler, seed, ...) embedded in PNG files are appended to each analysis result.",
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.includeGenerationMetadata
  )
  .field(
    "serverMaxTokens",
    "numeric",
    {
      displayName: "Vision API: Max Tokens",
      subtitle: "Maximum response length in tokens (1-4096). Default: 768.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.serverMaxTokens
  )
  .field(
    "serverTemperature",
    "numeric",
    {
      displayName: "Vision API: Temperature",
      subtitle: "Sampling temperature (0.0-2.0). Default: 0.7.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.serverTemperature
  )
  .field(
    "qwen3VlOdPrompt",
    "string",
    {
      displayName: "Qwen3-VL: Object Detection Prompt",
      subtitle: "Instruction sent to Qwen3-VL for default object detection. Leave empty to use the built-in default.",
      placeholder: "",
      isParagraph: true,
      engineDoesNotSupport: false,
    },
    defaultPluginSettings.qwen3VlOdPrompt
  )
  .field(
    "detectMaxTokens",
    "numeric",
    {
      displayName: "Vision API Detect: Max Tokens",
      subtitle: "Maximum response length in tokens for object detection (1-4096). Default: 2048.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.detectMaxTokens
  )
  .field(
    "detectTemperature",
    "numeric",
    {
      displayName: "Vision API Detect: Temperature",
      subtitle: "Sampling temperature for object detection (0.0-2.0). Default: 0.3.",
      engineDoesNotSupport: true,
    },
    defaultPluginSettings.detectTemperature
  )
  .build();