Task-aware model selection
A complexity classifier scores a coding turn in single-digit milliseconds and helps determine which model is appropriate for the task.
Weave Router는 코딩 에이전트를 위한 지능형 모델 라우터입니다. 코딩 작업을 평가해 오픈 모델과 클로즈드 모델로 요청을 라우팅하므로, 개발자는 기존 에이전트 워크플로를 유지하면서 속도, 품질, 모델 비용을 관리할 수 있습니다.
Weave Router is a model-routing layer for coding agents. It reads the turns an agent takes, assesses the task, and sends each request to a model selected for the job rather than using one model for every interaction. The product is designed to preserve existing coding-agent workflows while balancing task quality, response time, and provider cost.
The router supports workflows involving Claude Code, Codex, Cursor, OpenCode, and Pi, depending on the plan or setup. Users can add the model-provider keys or subscriptions they want to route across. Weave presents Router as a way to use lower-cost models for suitable work and reserve frontier models for tasks that need them.
A complexity classifier scores a coding turn in single-digit milliseconds and helps determine which model is appropriate for the task.
Router considers whether switching models would cost more than the expected saving from routing the request to a less expensive option.
An escalation classifier monitors task progress and can return a stalled, looping, or unsuccessful task to a frontier model.
The routing pool can include open and closed models, with examples on the site including DeepSeek, GLM, Kimi, Muse, Gemini, Claude, GPT-5, and Llama.
A single command is provided to detect and configure supported coding tools: `npx @workweave/router claude code codex cursor`.
The product provides comparisons using coding benchmarks and an estimator for projected monthly spend. Results are described as dependent on workload, prompt mix, and provider pricing.
Route routine tasks such as explanations, release-note drafts, configuration changes, or ticket triage to less expensive models while reserving stronger models for debugging and complex refactors.
Use Code Boost to route across existing subscriptions and additional models rather than replacing the agent interface or changing the coding workflow.
Use Code Max as a separate subscription built around tuned open-source models and Router-based selection.
Compare Router with other model-selection approaches using task success, elapsed time, and cost per trial on the listed Terminal-Bench and SWE-Atlas evaluations.
Organizations that need Router in their own infrastructure can discuss a custom Enterprise deployment and hands-on setup support.
The site names Claude Code, Codex, Cursor, OpenCode, and Pi. The home page also provides a setup command for Claude Code, Codex, and Cursor.
No. The product is presented as working with existing agents. Code Boost is specifically designed to extend Claude and Codex subscriptions, while Code Max is positioned as a separate subscription for open-source-model coding.
Router sends the coding request to a selected model within the configured pool. The site illustrates routing decisions with the selected model, a score, and an indicative cost level; actual results depend on the selected providers and task.
Router is billed separately from Weave Platform. The listed plans are Code Max at $10 per month, Code Boost at $100 per month, and custom Enterprise pricing for on-premises deployment. The source pages do not specify general usage limits.
트래픽 데이터는 참고용으로만 확인하세요.