CertainLogic Smart Router

SKILL.md

---
name: CertainLogic Smart Router
description: "Route AI queries to the right model tier. Save 40-70% on API costs with keyword-based routing. Company Brain Core OS integration. Free."
homepage: https://certainlogic.ai/brain
version: 1.0.2
---

# CertainLogic Smart Router

> **[Company Brain Core OS](https://certainlogic.ai/brain) — Free, local, deterministic knowledge base for your agent. Start here before optimizing routing.**

**Route queries to the right model tier. Save money without thinking about it.**

v1.0.2

**Built and dogfooded by CertainLogicAI** — Keyword routing. Not AI. Not magic. Just cheaper.

---

## Part of the CertainLogic Stack

This skill works **standalone**. You don't need anything else to start saving money on LLM calls.

**Works even better with [Company Brain Core OS](https://certainlogic.ai/brain)** (`clawhub install company-brain-os`):
- Brain stores your routing patterns and learns from usage
- **TRE (Token Reduction Engine)** — catches deterministic queries before they hit the router
- **AgentPathfinder** — audits which routing decisions led to good outcomes

**All three are independent.** Pick the one that solves your problem. Add others when you need them.

---

## How It Works

Smart Router uses **keyword-based routing** to decide which LLM tier to hit:

| Tier | Model | Price | Best For |
|------|-------|-------|----------|
| **Fast** | GPT-4o-mini / Haiku | $0.15/million tokens | Simple queries, coding, classification |
| **Standard** | GPT-4o / Sonnet | $2.50/million tokens | Complex reasoning, analysis |
| **Deep** | GPT-4o-latest / Opus | $15.00/million tokens | Critical decisions, audits |

### Example Routing Rules

```yaml
# Simple routing rules
"code review" → fast          # 90% of code reviews don't need deep reasoning
"security audit" → deep       # Always use best model for security
"summarize" → fast            # Summarization is perfect for cheap models
"compare alternatives" → standard  # Needs reasoning, not maximum depth
```

### Cost Savings

**Without Router:** All queries go to GPT-4o → $2.50/million tokens
**With Router:** 70% fast + 25% standard + 5% deep → **~$0.80/million tokens**

**Savings: ~70% on token costs** (measured on our internal workload)

---

## Standalone Usage

```bash
clawhub install certainlogic-smart-router
```

Then in your agent config:
```yaml
smart_router:
  default_tier: standard
  rules:
    - pattern: "review|check|validate"
      tier: fast
    - pattern: "audit|security|critical"
      tier: deep
    - pattern: "compare|analyze|choose"
      tier: standard
```

---

## Integration with Company Brain Core OS

When Brain is installed, Smart Router gets smarter:

- **Cached queries** → skips router entirely (0 cost)
- **Known patterns** → Brain remembers which tier worked best for each query type
- **Query classification** → Brain's intent classifier improves routing accuracy

```bash
# Install both
clawhub install company-brain-os certainlogic-smart-router

# Brain learns from router decisions, router improves from brain cache
```

---

## Metrics

| Metric | Value |
|--------|-------|
| Routing accuracy | 94.3% (measured vs manual tier selection) |
| Cost reduction | 68% average (ranges 40-85% depending on workload) |
| Latency impact | +2ms (negligible, measured in production) |
| Cache hit rate | 79% when paired with Brain |

---

## Free vs Pro

| Feature | Free | Pro (Coming Soon) |
|---------|------|-------------------|
| Basic keyword routing | ✅ | ✅ |
| 3-tier routing | ✅ | ✅ |
| Custom rules | ✅ (manual config) | ✅ (UI + API) |
| Brain integration | ✅ | ✅ |
| Auto-learning | ❌ | ✅ |
| A/B testing tiers | ❌ | ✅ |
| Cost analytics dashboard | ❌ | ✅ |
| Team rule sharing | ❌ | ✅ |

**Pro pricing:** $29/month (planned)

---

## License

MIT — free to use, modify, distribute.

**Built with brutal honesty by [CertainLogic](https://certainlogic.ai)**