Configure AI Providers
Connect your AI model to enable Explain Findings and NL Policy Generation. SOFE never hosts or pays for inference — you bring your own model.
Why You Need This
AI features (Explain, Generate Policy) require a language model. SOFE routes your request to YOUR configured provider. Your API key, your costs, your data.
Quick Start: OpenRouter (Free, 2 Minutes)
- 1. Go to openrouter.ai/keys → Sign in with Google → Create key
- 2. In SOFE: Settings → AI Provider → Select "OpenRouter"
- 3. Paste your API key → Save AI Config
- 4. Select a model (free models marked with 🆓) → Save
- 5. Click "Test Connection" → Should say ✅ Connected
All Supported Providers
| Provider | Free Tier? | Auth | Best For |
|---|---|---|---|
| OpenRouter | ✅ Free models | API key | Quick start, multiple models |
| NVIDIA NIM | ✅ 1000 free calls | API key | Fast inference, GPU-optimized |
| HuggingFace | ✅ Free inference | API key | Open models, testing |
| Amazon Bedrock | Pay per use | AWS IAM role | Enterprise, privacy, no key needed |
| OpenAI | Pay per use | API key | Highest quality (GPT-4o) |
| Databricks | Workspace trial | API key + URL | MLOps teams, DBRX |
| Huawei ModelArts | Trial credits | API key | China market, Pangu models |
Amazon Bedrock (No API Key)
If you've connected your AWS account via CloudFormation, Bedrock works automatically — no API key needed. Your IAM role just needs bedrock:InvokeModel permission (included in the latest CF template).
# If you deployed the CF stack before Jul 7, re-deploy to get Bedrock permissions: # Go to platform.sofe.dev/accounts → Re-launch CloudFormation stack
Rate Limits (AI Calls)
| Tier | AI Calls/Day | Note |
|---|---|---|
| Free | 5 | Proof of concept |
| Pro | 100 | Daily review |
| Enterprise | 1000 | Bulk analysis |
AI calls are click-only — never auto-triggered. Each "Explain" or "Generate Policy" click counts as 1 call.
Troubleshooting
❌ "No AI provider configured" → Go to Settings → AI Provider → select a provider and save
❌ "Provider error (429)" → Your model is rate-limited upstream. SOFE auto-retries with a fallback model. Or switch to a different model in Settings.
❌ "AI rate limit exceeded" → You've used all daily AI calls. Resets at midnight UTC. Upgrade for more.
❌ "Invalid API key" → Re-check your key in Settings. OpenRouter keys start with
sk-or-v1-Related: AI Explain Findings • Create Policies with AI