The "model-routing" topic on GitHub groups 27 open-source projects in the DeepSeek Harness (DSH) ecosystem, led by router-core with 10 GitHub stars. router-core — Smart LLM router for AI agents — automatic model selection on every request: classify, hard-filter on capability, then rank on task fit, cost, speed and reliability. Every project here is indexed by DSH Universe with live GitHub data — stars, activity and install status — so you can compare and install directly.
Smart LLM router for AI agents — automatic model selection on every request: classify, hard-filter on capability, then rank on task fit, cost, speed and reliability. Deterministic and local, <1ms, no inference call. Powers ClawRouter, Franklin, Hermes and dsh-clawrouter.
🌊 kimi-tide — DSH's "auto-select a model at every step" plugin: images go to a model that can see them, coding goes to a coding model, chit-chat and translation go to a cheap model; the panel shows who was picked and why each time · The per-step model router for DeepSeek Harness: any connected model, routed by your rules
Local-model (Ollama) integration for DeepSeek Harness: discover, pull, remove, and inspect local models, route requests to them by task type or keyword with automatic fallback to the cloud, and get a one-shot status overview via /ollama.
DSH subagent router: per-role routing with candidate pools, cross-provider failover, fork window slicing, mid-run and batch model switching — plus header critters showing who is running, for how long and on how many tokens, and model-callable routing tools.
Model auto-ranking and transparent failover for DeepSeek Harness: ranks discovered models and switches to the next candidate when the current one fails with no output.
Automatic model routing for DeepSeek Harness: three difficulty tiers (hard/normal/easy) plus vision routing, picking models you already configured under Settings ? Models. ???? + ?????????
DeepSeek Harness plugin: role-based model routing — planner (root agent) on deepseek-v4-pro, delegated executor subagents on deepseek-v4-flash; ships a prompt section and a pro-flash-routing skill.
Tiered model routing with evidence-based disqualification: a harness-neutral skill, a Rust engine and CLI, a hosted MCP server (Cloudflare Worker in Rust) and a dsh adapter.
DeepSeek Harness plugin: pick the model your subagents run on — with a reasoning-effort ceiling and an output-token cap — enforced in code on every child DSH starts, whichever tool starts it.
A routing plugin for DeepSeek Harness that assigns models by task: a delegated subagent runs on the model that fits its job, while the session you are talking to always keeps the model you picked. | 为 DeepSeek Harness 提供按任务分配模型的路由插件: 被委派出去的子智能体跑在适合它那件事的模型上,而你正在对话的这个会话,用的始终是你选的那个模型。
DSH plugin: subagents default to a cheaper sibling of the parent model (Opus to Sonnet, Qwen max to flash, DeepSeek pro to flash) unless the agent names a route
Model routing for DeepSeek Harness, with a standing capability-to-model assignment table, per-route reasoning levels, and a user-triggered sync that researches public model prices. 为 DeepSeek Harness 提供模型路由:可持久化的「能力→模型」分工表、按路由的推理档位,以及手动触发、由模型联网核对公开价格的同步。