lynkr-dsh-plugin

Juggling Anthropic, OpenAI, Bedrock, Ollama and a dozen other providers inside DeepSeek Harness usually means one of two things: you are paying top-tier prices for trivial completions, or you are under-pushing cheap models on reasoning-heavy tasks. lynkr-dsh-plugin registers a self-hosted Lynkr gateway as a custom OpenAI-compatible provider in dsh. Lynkr sits in front of your coding agents and chat clients, classifies the difficulty of every incoming request, and routes it to the cheapest model tier across 15+ providers that can actually handle the job. The model name you see in dsh's picker is just a label; Lynkr's server-side tier router makes the real routing decision.

The plugin is deliberately conservative in how it touches your configuration. It never overwrites an existing lynkr entry and never reaches into another provider's settings. If anything about the automatic merge or the plugin-loading path does not behave as expected, it fails safe and prints the exact YAML block you can paste in by hand, so there is always a loader-independent fallback that works.

This is for developers who already run multiple LLM providers through dsh, want to replace manual model switching with per-request automated tier selection, and prefer keeping the routing gateway under their own control. Spin up a Lynkr instance, pick the lynkr provider in dsh, and let its tier router decide which backend server actually serves each request based on complexity.