StrataRouter

AI Execution Control Plane

Production-grade semantic routing for AI systems. Fast Rust core. Hybrid scoring. 9 framework integrations.

---

What is StrataRouter?

StrataRouter is an AI Execution Control Plane — infrastructure-level routing for production AI systems.

While LangChain and LlamaIndex provide prompt-level routing helpers, StrataRouter provides deterministic execution, cost-aware model selection, governance, and compliance at the system level.

graph TD
    Q["Your Query"] --> CORE

    subgraph CORE["StrataRouter Core · MIT · this repo"]
        A["Intent Detection"] --> B["Hybrid Scoring\n(dense · BM25 · rule)"]
        B --> C["Confidence Calibration"]
    end

    CORE --> RUNTIME

    subgraph RUNTIME["StrataRouter Runtime · Apache 2.0"]
        D["TCFP Execution"] --> E["Semantic Cache"]
        E --> F["REST API · Prometheus"]
    end

    RUNTIME --> ENTERPRISE

    subgraph ENTERPRISE["StrataRouter Enterprise · Commercial"]
        G["Consensus Engine"] --> H["Immutable Audit Log"]
        H --> I["Policy Engine · Multi-tenant"]
    end

Loading

---

Why StrataRouter?

Most AI frameworks treat routing as an afterthought — a few if/else chains or a basic vector similarity lookup. StrataRouter treats routing as infrastructure:

Problem	StrataRouter Answer
Slow Python similarity loops	Rust core — 8.7 ms P99 latency
Memory bloat at scale	64 MB for 1K routes (vs 2.1 GB in pure-Python routers)
No confidence calibration	Per-route piecewise-linear score normalisation
No hybrid keyword + semantic	BM25 + dense embeddings combined
No audit trail	Immutable SHA-256 log (Enterprise)
No cost control	Per-route budget enforcement (Enterprise)

---

⚡ Benchmarks

vs semantic-router

Metric	StrataRouter	semantic-router	Delta
P99 Latency	8.7 ms	178 ms	~20× faster
Memory (1K routes)	64 MB	2.1 GB	~33× less
Throughput	18K req/s	450 req/s	~40× higher
Accuracy	95.4%	84.7%	+12.7%

Benchmarks run on Ubuntu 22.04, AMD EPYC 7B13, Python 3.11, sentence-transformers/all-MiniLM-L6-v2. See benchmarks/ for methodology and reproduction scripts.

vs Competitors

Feature	StrataRouter	semantic-router	LangChain Router	route0x
Rust core	✅	❌	❌	❌
Hybrid BM25 + dense	✅	❌	❌	❌
Confidence calibration	✅	❌	❌	❌
Cost-aware routing	✅	❌	❌	❌
Audit log	✅	❌	❌	❌
Sub-10ms P99	✅	❌	❌	❌

---

📦 Installation

pip install stratarouter
pip install stratarouter[huggingface]   # local embeddings, no API key
pip install stratarouter[openai]        # OpenAI embeddings
pip install stratarouter[cohere]        # Cohere embeddings
pip install stratarouter[all]           # everything

---

🚀 Quick Start

High-level API — RouteLayer (recommended)

from stratarouter import Route, RouteLayer
from stratarouter.encoders import HuggingFaceEncoder

routes = [
    Route(
        name="billing",
        utterances=["invoice", "payment", "refund", "charge"],
        threshold=0.75,
    ),
    Route(
        name="support",
        utterances=["help", "broken", "error", "can't login"],
        threshold=0.75,
    ),
]

encoder = HuggingFaceEncoder()
rl = RouteLayer(encoder=encoder, routes=routes)

result = rl("I need my April invoice")
print(result.name)       # "billing"
print(result.score)      # 0.87
print(bool(result))      # True — score >= threshold

Low-level API — Router (Rust core, advanced use)

from stratarouter import Router, Route

router = Router(encoder="sentence-transformers/all-MiniLM-L6-v2")
router.add(Route(name="billing", utterances=["Where's my invoice?", "I need a refund"]))
router.add(Route(name="support", utterances=["App is crashing", "Can't login"]))
router.build_index()

result = router.route("I need my April invoice")
print(result.route_id)    # "billing"
print(result.confidence)  # 0.89 — calibrated score
print(result.latency_ms)  # 2.3

router.save("my_router.json")
router = Router.load("my_router.json")   # no re-indexing needed

---

🏗️ Routing Pipeline

flowchart LR
    Q["User Query"] --> E["Embedding\nEncoder"]
    E --> HS

    subgraph HS["Hybrid Scorer · Rust"]
        D["Dense\nCosine\n×0.6427"]
        B["BM25\nKeyword\n×0.2891"]
        R["Rule\nPattern\n×0.0682"]
    end

    D --> CAL["Calibration\nper-route\npiecewise-linear"]
    B --> CAL
    R --> CAL
    CAL --> SEL["Route\nSelection\nargmax"]
    SEL --> OUT["RouteResult\nroute_id · confidence\nlatency_ms"]

Loading

---

🎨 Framework Integrations (9)

from stratarouter.integrations.langchain         import StrataRouterChain
from stratarouter.integrations.langgraph         import create_routing_graph
from stratarouter.integrations.crewai            import RoutedAgent
from stratarouter.integrations.autogen           import StrataRouterGroupChat
from stratarouter.integrations.openai_assistants import StrataRouterAssistant
from stratarouter.integrations.google_agent      import StrataRouterVertexAI

Runnable examples → integrations/

---

🗺️ Roadmap

Version	Feature	Status
v0.1	Prototype — RouteLayer API	✅ Nov 2025
v0.2	Production Rust engine, 9 integrations	✅ Mar 2026
v0.3	Runtime — TCFP, REST API, cache	✅ Available
v0.4	Cost optimizer — model selection, budgets	🔄 In Progress
v0.5	Enterprise governance	✅ Available (private)
v0.6	JS / Go SDKs	📋 Q3 2026
v1.0	StrataRouter Cloud	📋 Q4 2026

Full roadmap → ROADMAP.md

---

📚 Documentation

Getting Started	Install + first router in 5 min
API Reference	RouteLayer, Router, Route, RouteChoice
Integrations	All 9 framework guides
Architecture	Rust internals, scoring, calibration
Deployment	Docker, K8s, server
Changelog	Release history
Roadmap	What's coming

Full docs → docs.stratarouter.com

---

🏗️ Development

git clone https://github.com/ai-deeptech/stratarouter.git
cd stratarouter

pip install -e "python/.[dev]"
cd python && maturin develop --release && cd ..

make test

---

🏢 Platform Tiers

	Core (MIT)	Runtime (Apache 2.0)	Enterprise
Semantic routing library	✅	✅	✅
9 framework integrations	✅	✅	✅
PyPI package	✅	—	—
TCFP workflow execution	—	✅	✅
REST API + Prometheus	—	✅	✅
Semantic cache + batch dedup	—	✅	✅
Multi-agent consensus	—	—	✅
Immutable audit log (SOC2/HIPAA)	—	—	✅
Policy engine (RBAC/ABAC)	—	—	✅
Multi-tenant isolation	—	—	✅

→ Runtime
→ Enterprise: support@stratarouter.com
→ Docs: docs.stratarouter.com
→ Website: stratarouter.com

---

🤝 Contributing

See CONTRIBUTING.md · SUPPORT.md · CODE_OF_CONDUCT.md

---

📝 License

MIT — LICENSE.txt
Built with PyO3 · Sentence Transformers
Made with ⚡ by StrataRouter Contributors