EAX Router ALPHA
An experimental, intelligent, and pluggable routing layer for Large Language Models. Research into dynamic model selection based on cost, latency, or quality priorities.
⚠️ Experimental Research - This project explores intelligent routing concepts and will remain permanently in alpha status.
🧠 Intelligent Model Selection
Research into how applications can dynamically choose the optimal LLM for each task based on priority settings - cost, latency, or quality optimization.
🔌 Pluggable Routing Strategies
Experimental framework for different routing approaches: CostOptimized, LatencyOptimized, QualityOptimized, with interfaces for custom strategies.
📊 Model Fingerprinting
Research into standardized model capability descriptions through fingerprint.json files, enabling community-driven benchmarking and routing intelligence.
🌐 Provider Agnostic
Experimental design to work across any LLM provider - OpenAI, Anthropic, Google, or open-source models via Ollama and other platforms.
⚡ Lightweight Design
Research into minimal overhead routing that adds intelligence without significant latency or complexity to existing application stacks.
🔬 Open Research
Community-driven model performance data collection and sharing, making routing decisions smarter through collaborative benchmarking efforts.
Research Hypothesis
Can a smart routing layer enable developers to use the right model for the right job automatically, reducing costs and vendor lock-in?
from eax_router import Router, Task
# 1. Initialize the experimental router
router = Router()
# 2. Define your task
task = Task(
prompt="Summarize this 500-word article into three bullet points.",
priority="cost" # or "latency", or "quality"
)
# 3. Intelligent routing!
model_choice = router.route(task)
# >>> Model chosen: 'claude-3-haiku-20240307' (cost-optimized)
Fingerprint Standard Research
Exploring an open standard for LLM capability description to enable intelligent routing decisions.
{
"model_id": "claude-3-haiku-20240307",
"cost_per_million_tokens": { "input": 0.25, "output": 1.25 },
"avg_latency_ms_per_1k_tokens": 350,
"context_window": 200000,
"performance_by_task": {
"summarization": { "quality_score": 8.5 },
"code_generation": { "quality_score": 6.0 }
}
}