AI Explain Findings
Get instant AI-powered explanations for any finding — why it matters, what's wrong, and how to fix it.
How It Works
Click "🧠 Explain with AI" on any finding card. SOFE sends the finding context to your configured AI provider and returns a structured explanation.
Click "Explain" → SOFE builds prompt with finding context
→ Calls your AI provider (ai-api.sofe.dev/explain)
→ Returns structured response:
**Why this matters:** (impact)
**What's wrong:** (specific to this resource)
**Fix:** (actionable steps)
**Estimated impact:** (savings/risk reduction)Prerequisites
- AI Provider configured in Settings → AI Provider
- At least 1 evaluation run (findings to explain)
- AI rate limit not exceeded (5/day free, 100/day pro)
Where to Find It
• /history → Click evaluation → Finding card → "🧠 Explain with AI"
• /topology → Click node → Finding in sidebar → "🧠 Explain with AI"
Example Output
Why this matters: Untracked resources lead to inaccurate cost allocation and obscure financial accountability.
What's wrong: The Lambda function htmlprueba in us-east-1 is missing the required owner tag.
Fix:
- Add the
ownertag with the responsible team's identifier - Update your IaC to enforce this tag on future deployments
Estimated impact: Ensures accurate chargeback reporting and prevents unallocated costs.
🏷️ openrouter/claude-sonnet-5 • 4.5s
Model Recommendations
| Model | Speed | Quality | Cost |
|---|---|---|---|
| Claude Sonnet 5 | ~4s | ⭐⭐⭐⭐⭐ | ~$3/M tokens |
| Llama 3.3 70B (free) | ~6s | ⭐⭐⭐⭐ | Free |
| Gemma 4 31B (free) | ~8s | ⭐⭐⭐ | Free |
| GPT-4o Mini | ~3s | ⭐⭐⭐⭐ | ~$0.15/M tokens |
Models under 30B parameters (nano, mini, 7B, 8B) are filtered out — they can't reliably follow structured output instructions.
Important Notes
- Click-only: Explain is never triggered automatically. Each click = 1 AI call.
- Your costs: SOFE doesn't pay for inference. Costs go to YOUR provider account.
- Rate limited separately: AI calls have their own limit (5/100/1000), independent of evaluations.
- 429 fallback: If your model is rate-limited, SOFE auto-retries with an alternative free model.