CertainLogic Smart Router

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

v1.0.0

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

What It Actually Does

A deterministic keyword + regex router that classifies your query and recommends the best LLM tier:

cheap — Simple lookups, greetings, short answers
default — Standard tasks, explanations, drafting
powerful — Complex reasoning, architecture, strategy

It supports built-in profiles (coding, research, marketing, general) and custom profiles via JSON.

What It Does NOT Do

Claimed Feature	Reality
AI-powered or learned routing	NO — Pure keyword + regex matching (deterministic)
Automatic LLM API calls	NO — Returns recommendation only. You call the model.
Built-in caching (v1)	NO — Caching layer planned for v1.1
Guaranteed perfect routing	NO — Simple heuristics. Edge cases may misroute.
Multi-provider auto-selection	NO — Works with any models (you map tiers to models)

How to Use

Standalone CLI

python3 scripts/smart_router.py "Write a Python function to parse JSON"

In Your Agent (example)

from smart_router import SmartRouter

router = SmartRouter()
result = router.route("Design a distributed system")

if result["model_tier"] == "cheap":
    model = "anthropic/claude-haiku-4-5"
elif result["model_tier"] == "default":
    model = "anthropic/claude-sonnet-4-6"
else:
    model = "anthropic/claude-opus-4-6"

# Then call your LLM with the chosen model

Override Flags

--cheap — Force cheap tier
--powerful — Force powerful tier
--config my_profiles.json — Use custom profiles

Honest Limitations

Static keyword lists (update config for new domains)
English only (patterns tuned for English)
Returns recommendation only (no API calls)
Simple heuristics — will misroute edge cases (override flags exist)
No built-in usage analytics or dynamic learning in v1

Free vs Pro

Free (this skill)

4 built-in profiles
Custom config support
Override flags
Full source code on GitHub

Pro ($29 one-time)

Dynamic feedback loop (track what actually worked)
Usage analytics + monthly savings report
Auto-optimization using our benchmarks
Fallback chains and team profiles
Priority support

Example Savings

Scenario: 100 queries/day (estimates)

Without Router	With Router
100 × Opus	60 × Haiku + 30 × Sonnet + 10 × Opus
~$150/day	~$120/day
	Savings: ~20%

Note: Actual savings depend on your query mix. Test with our benchmarks.

Recommended Next Steps (CertainLogic Stack)

Skill Vetter Plus — Scan before installing anything new
Token Reduction Engine — Keep sessions lean and cheap
AgentPathfinder — Verifiable task tracking
Skill Oracle — Honest skill recommendations

All work great together.

Open Source Core

Full source: https://github.com/CertainLogicAI/certainlogic-smart-router
Related benchmarks: https://github.com/CertainLogicAI/hallucination-benchmark