Weave Router intelligently directs coding agent requests to optimal LLMs, cutting costs 40-70% without sacrificing quality. The system automatically selects between Opus 4.8, GPT 5.5, and cheaper models like DeepSeek V4 based on task complexity. By analyzing tens of thousands of agent traces through reinforcement learning, it routes critical tasks to frontier models while using efficient alternatives for routine operations. Self-hosted via Elastic License 2.0, it acts as a drop-in proxy for Anthropic, OpenAI, and Gemini APIs.

Developers integrate it through simple configuration changes, with localhost:8080 handling all routing decisions and providing observable metrics for cost optimization.