helicone / rules

---
description: 
globs: docs/ai-gateway/*
alwaysApply: false
---
# AI Gateway Documentation Rules

Critical rules to ensure documentation accuracy based on configuration changes.

## Configuration Structure (BREAKING CHANGES)

### 1. Cache Store Configuration

```yaml
# ✅ CORRECT - cache-store at top level
cache-store:
  type: "in-memory"

routers:
  my-router:
    cache:
      directive: "max-age=3600"

# ❌ WRONG - old structure
routers:
  my-router:
    cache:
      store: in-memory
      enabled: true
```

### 2. Helicone Configuration

```yaml
# ✅ CORRECT - separate fields
helicone:
  authentication: true
  observability: true

# ❌ WRONG - old structure
helicone-observability:
  enable-auth: true
```

## Required Dependencies

### 3. Authentication Requirements

- **Rate limiting requires authentication** (works per-API-key)
- **Observability requires authentication** to function
- Always add warning and link to `/ai-gateway/authentication`

### 4. Model Names

```yaml
# ✅ CORRECT - provider prefix
"model": "openai/gpt-4o-mini"

# ❌ WRONG - no prefix
"model": "gpt-4o-mini"
```

## Examples That Actually Work

### 5. Code Examples

- **Always include:** TypeScript, Python, cURL (no Node.js)
- **Router URLs:** `http://localhost:8080/router/{name}/v1/chat/completions`
- **API Keys:** Use `sk-placeholder` for local, real keys for public deployments

### 6. Deprecated Patterns to Never Use

- ❌ `cache.enabled: true/false`
- ❌ `cache.store: in-memory`
- ❌ `helicone-observability.enable-auth`
- ❌ Model names without provider prefixes

More agent context in Helicone/helicone

7 other files this repository gives its agents.