OpenAgentsControl/.opencode/tool/gemini-multiple.ts

import { tool } from "@opencode-ai/plugin"

async function parseImageInput(input: string) {
  // Accepts file path ("./img.png") or data URL ("data:image/png;base64,...")
  if (input.startsWith("data:")) {
    const base64 = input.split(",")[1]
    const mime = input.substring(5, input.indexOf(";"))
    return { mime, base64 }
  }
  // Treat as file path
  const file = Bun.file(input)
  const arr = await file.arrayBuffer()
  const base64 = Buffer.from(arr).toString("base64")
  // Best-effort mime
  const mime = file.type || "image/png"
  return { mime, base64 }
}

async function callGeminiAPI(mime: string, base64: string, prompt: string) {
  const apiKey = process.env.GEMINI_API_KEY
  if (!apiKey) throw new Error("Set GEMINI_API_KEY in your environment")

  const res = await fetch(
    "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image-preview:generateContent",
    {
      method: "POST",
      headers: {
        "Content-Type": "application/json",
        "x-goog-api-key": apiKey,
      },
      body: JSON.stringify({
        inputs: [{ mimeType: mime, data: base64 }],
        contents: [{ parts: [{ text: prompt }]}],
      }),
    }
  )

  if (!res.ok) throw new Error(`API error: ${await res.text()}`)
  const json = await res.json()
  const b64 = json?.candidates?.[0]?.content?.parts?.[0]?.inlineData?.data
  if (!b64) throw new Error("No image data returned")
  
  return b64
}

export const edit = tool({
  description: "Edit an image using Gemini with file path or data URL",
  args: {
    image: tool.schema.string().describe("File path or data URL"),
    prompt: tool.schema.string().describe("Edit instruction"),
    output: tool.schema.string().optional().describe("Output filename (default edited.png)"),
  },
  async execute(args, context) {
    try {
      const { mime, base64 } = await parseImageInput(args.image)
      const resultBase64 = await callGeminiAPI(mime, base64, args.prompt)
      
      const out = args.output || "edited.png"
      await Bun.write(out, Buffer.from(resultBase64, "base64"))
      return `Saved edited image to ${out}`
    } catch (error) {
      return `Error: ${error.message}`
    }
  },
})

export const analyze = tool({
  description: "Analyze an image using Gemini without editing",
  args: {
    image: tool.schema.string().describe("File path or data URL"),
    question: tool.schema.string().describe("What to analyze about the image"),
  },
  async execute(args, context) {
    try {
      const { mime, base64 } = await parseImageInput(args.image)
      
      const apiKey = process.env.GEMINI_API_KEY
      if (!apiKey) return "Set GEMINI_API_KEY in your environment"

      const res = await fetch(
        "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent",
        {
          method: "POST",
          headers: {
            "Content-Type": "application/json",
            "x-goog-api-key": apiKey,
          },
          body: JSON.stringify({
            contents: [{
              parts: [
                { text: args.question },
                { inlineData: { mimeType: mime, data: base64 } }
              ]
            }],
          }),
        }
      )

      if (!res.ok) return `API error: ${await res.text()}`
      const json = await res.json()
      const text = json?.candidates?.[0]?.content?.parts?.[0]?.text
      if (!text) return "No analysis returned"
      
      return text
    } catch (error) {
      return `Error: ${error.message}`
    }
  },
})