Weave Router 2.0 is an intelligent model router for coding agents that evaluates turn complexity, manages prompt cache efficiency, and routes requests across frontier and alternative models.
What the product does and how it is positioned
Weave Router 2.0 is designed for coding agents to direct individual turns between different language models based on prompt complexity and context requirements.
The router integrates with development environments such as Claude Code, Codex, and Cursor, directing queries across model families including Claude, GPT, DeepSeek, GLM, and Gemini.
Source-supported ways to use the product
Routing automated terminal command tasks through complexity classifiers to balance execution between frontier and alternative models.
Managing multi-turn codebase analysis sessions while preserving prompt caching efficiency across model switches.
The documented workflow, where available
Execute the router initialization command to scan and configure supported tools such as Claude Code, Codex, and Cursor.
Supply the API keys and active subscription credentials across the targeted model providers.
Direct coding agent requests through the router to dynamically balance turns across the configured model pool.
Weave Router 2.0 evaluates agent requests through a millisecond-level classification pipeline. The complexity classifier determines turn difficulty before dispatching requests, taking into account the state and cost of rebuilding prompt caches before routing down to secondary models.
During task execution, an integrated escalation classifier continuously monitors model output. If a delegated model encounters an execution loop, stalls, or fails to complete required steps, the router intervenes and escalates the turn back to a frontier model.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
The router setup utility automatically detects and configures Claude Code, Codex, and Cursor coding harnesses.
Weave Router 2.0 supports routing across multiple model families, including Claude, GPT, DeepSeek, GLM, and Gemini.
It evaluates turn complexity in single-digit milliseconds and confirms that switching will not incur excessive cache rebuild overhead.
An escalation classifier monitors for stalls, loops, or misses, automatically redirecting the request back to a frontier model.
The router identifies active flat-rate seats and priority subscriptions to utilize owned quota before dispatching requests elsewhere.
