arela/dist/analysis/vision.js

AI-powered CTO with multi-agent orchestration, code summarization, visual testing (web + mobile) for blazing fast development.

// src/analysis/vision.ts
import { exec } from 'child_process';
import { promisify } from 'util';
const execAsync = promisify(exec);
export async function analyzeWithMoondream(screenshotPath, context) {
    // Check if Ollama is available
    try {
        await execAsync('ollama --version');
    }
    catch {
        // Gracefully return empty result instead of throwing
        console.log('\n⚠️  Ollama not installed - AI analysis unavailable');
        console.log('💡 Install Ollama for FREE AI-powered analysis:');
        console.log('   brew install ollama');
        console.log('   (Falling back to rule-based checks only)\n');
        return {
            issues: [],
            summary: 'AI analysis unavailable (Ollama not installed)',
        };
    }
    // Check if moondream is available
    try {
        await execAsync('ollama list | grep moondream');
    }
    catch {
        console.log('📥 Pulling moondream model (first time only, ~1GB)...');
        try {
            await execAsync('ollama pull moondream');
            console.log('✅ Moondream model ready!\n');
        }
        catch (error) {
            console.log('\n⚠️  Failed to pull moondream model');
            console.log('💡 Try manually: ollama pull moondream');
            console.log('   (Falling back to rule-based checks only)\n');
            return {
                issues: [],
                summary: 'AI analysis unavailable (failed to pull model)',
            };
        }
    }
    // Prepare prompt
    const prompt = `You are analyzing a screenshot of a web application for UX and accessibility issues.

Context: ${context}

Analyze this screenshot and identify:
1. UX Issues - Poor contrast, small text, unclear buttons, missing feedback
2. Accessibility Issues - WCAG violations, small touch targets, poor color contrast
3. Visual Issues - Broken layouts, overlapping elements, cut-off text

For each issue, provide:
- Severity (critical/warning/info)
- Category (ux/accessibility/visual)
- Clear message
- Actionable suggestion

Format as JSON:
{
  "issues": [
    {
      "severity": "critical",
      "category": "accessibility",
      "message": "Text has very low contrast",
      "suggestion": "Increase text darkness or background lightness"
    }
  ],
  "summary": "Found 2 issues: 1 critical, 1 warning"
}`;
    // Call Ollama with moondream
    try {
        const { stdout } = await execAsync(`ollama run moondream "${prompt}" < "${screenshotPath}"`);
        // Parse JSON from response
        const jsonMatch = stdout.match(/\{[\s\S]*\}/);
        if (!jsonMatch) {
            return {
                issues: [],
                summary: 'No issues detected by AI',
            };
        }
        try {
            const result = JSON.parse(jsonMatch[0]);
            return result;
        }
        catch {
            return {
                issues: [],
                summary: 'Failed to parse AI response',
            };
        }
    }
    catch (error) {
        console.log('\n⚠️  AI analysis failed');
        console.log('💡 Continuing with rule-based checks only\n');
        return {
            issues: [],
            summary: 'AI analysis failed (continuing with rules)',
        };
    }
}
//# sourceMappingURL=vision.js.map