hematite-cli 0.4.6

use crate::agent::inference::{ChatMessage, InferenceEngine};
use base64::prelude::*;
use serde_json::Value;
use std::path::Path;

pub fn encode_image_as_data_url(path: &Path) -> Result<String, String> {
    if !path.exists() {
        return Err(format!("File not found: {}", path.display()));
    }

    let data = std::fs::read(path).map_err(|e| format!("Failed to read image: {}", e))?;
    let b64 = BASE64_STANDARD.encode(data);

    let ext = path.extension().and_then(|e| e.to_str()).unwrap_or("png");
    let mime = match ext.to_lowercase().as_str() {
        "jpg" | "jpeg" => "image/jpeg",
        "gif" => "image/gif",
        "webp" => "image/webp",
        _ => "image/png",
    };

    Ok(format!("data:{};base64,{}", mime, b64))
}

pub async fn vision_analyze(engine: &InferenceEngine, args: &Value) -> Result<String, String> {
    let path_str = args
        .get("path")
        .and_then(|v| v.as_str())
        .ok_or("Missing parameter: path")?;
    let prompt = args
        .get("prompt")
        .and_then(|v| v.as_str())
        .ok_or("Missing parameter: prompt")?;

    let path = Path::new(path_str);
    let url = encode_image_as_data_url(path).map_err(|e| {
        if e.starts_with("File not found: ") {
            format!("File not found: {}", path_str)
        } else {
            e
        }
    })?;

    let messages = vec![
        ChatMessage::system("You are a vision-capable technical assistant. Analyze the provided image (likely a screenshot, diagram, or UI mockup) and provide a concise technical summary or answer the specific query."),
        ChatMessage::user_with_image(prompt, &url),
    ];

    // Use the main engine but with tools disabled for the vision-pass sub-call.
    let (text, _, _, _) = engine.call_with_tools(&messages, &[], None).await?;

    Ok(text.unwrap_or_else(|| "The vision model returned an empty response.".to_string()))
}