Rayline Labs
Rayline Labs provides ARC, a routing intelligence model that selects the optimal model and reasoning effort for each turn of an AI request, balancing success rate against cost and latency. ARC operates at the turn level, switching between models like DeepSeek and GPT within a single conversation, and is cache-aware to minimize input costs. Developers can use ARC through hosted generation endpoints or the Routes API for their own inference infrastructure.
- Artificial Intelligence
- Developer Tools
- AI Agents
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
AI agents and applications face a fragmented model ecosystem where no single model performs best across all tasks, costs, and latency requirements. Developers must manually choose models and reasoning configurations, often leading to suboptimal performance, excessive spend, or unnecessarily slow responses.
Solution
Rayline provides ARC, a routing intelligence model that predicts the most suitable model and reasoning effort for each turn of a request, optimizing for both success rate and total cost. ARC operates at the turn level, meaning it can switch between models like DeepSeek Non-thinking, DeepSeek Thinking, GPT xhigh, and Qwen Local dynamically within a single conversation. The ARC Decisions API recommends a model and reasoning configuration while the developer runs inference in their own stack, whereas the ARC Responses API combines routing and generation in a single request. The system is cache-aware, accounting for cached tokens to minimize input costs. Open weights for ARC are planned.
Target Audience
Primary customers are AI application developers, agent builders, and engineering teams using LLM infrastructure who need to optimize model selection, cost, and performance across heterogeneous model ecosystems.
Features
- Turn-level model selection that routes each step of an agent's execution to the best-suited model
- Cache-aware routing that reduces input costs by leveraging cached token rates
- ARC Decisions API for model selection with the developer's own inference infrastructure
- ARC Responses API that bundles routing and generation, with routing fees waived
- Agent Gateway built on vLLM Semantic Router with zero Rayline commission on model inference
- Runtime adapters for vLLM and llama.cpp for local deployment