RouterBase 模型路由
Use routerbase when an agent needs a practical model routing plan instead of a single hard-coded model. This skill guides the agent through matching workloads to model capabilities, defining fallback behavior, and documenting the tradeoffs that matter during production rollout: latency, cost, context length, tool calling support, structured output reliability, and media or multimodal capability.
The routing process starts with a workload inventory. Separate cheap classification, high-context analysis, tool-heavy agent loops, media generation, and final user-facing synthesis into different traffic classes. For each class, choose a primary model, one or two fallback models, retry limits, timeout budgets, and a logging plan that records route decisions without storing sensitive prompts or credentials. During validation, run a small prompt set through every route, compare output quality, verify error handling, and write down when the system should fail closed rather than silently degrading to a model that cannot support the requested feature.
Installation
OpenClaw
clawhub install routerbase-model-routing
Source
- routerbase
