Constant ADVANCED 

pub const ADVANCED: &str = r#"
## LLM-Powered Tools

These tools require an LLM provider configured in `~/.config/subcog/config.toml`.

### Memory Consolidation

Merge similar memories using LLM analysis:

```json
{
  "tool": "subcog_consolidate",
  "arguments": {
    "namespace": "learnings",
    "strategy": "merge",
    "dry_run": true
  }
}
```

**Strategies:**
- `merge` - Combine similar memories into one
- `summarize` - Create summary of related memories
- `dedupe` - Remove exact duplicates

### Memory Enrichment

Improve a memory with better structure and tags:

```json
{
  "tool": "subcog_enrich",
  "arguments": {
    "memory_id": "decisions_abc123",
    "enrich_tags": true,
    "enrich_structure": true,
    "add_context": true
  }
}
```

## LLM Provider Configuration

Configure in `~/.config/subcog/config.toml`:

```toml
[llm]
provider = "anthropic" # or "openai", "ollama", "lmstudio"
api_key = "${ANTHROPIC_API_KEY}"
model = "claude-3-haiku-20240307"
```

| Provider | Model | Use Case |
|----------|-------|----------|
| Anthropic | claude-3-* | Best quality |
| OpenAI | gpt-4o-mini | Fast, good quality |
| Ollama | llama3.2 | Local, private |
| LM Studio | varies | Local, flexible |

## Search Optimization

### Hybrid Search Tuning

For precision-focused results:

```json
{
  "tool": "subcog_recall",
  "arguments": {
    "query": "exact topic",
    "mode": "text",
    "limit": 5
  }
}
```

For concept-focused results:

```json
{
  "tool": "subcog_recall",
  "arguments": {
    "query": "general concept or idea",
    "mode": "vector",
    "limit": 10
  }
}
```

## Embedding Model

Default: `all-MiniLM-L6-v2` (384 dimensions)

The embedding model is used for semantic similarity search in vector mode.
"#;

Advanced features documentation.