Lynkr

Why we included this project

Lynkr is a local HTTP proxy that runs in front of your AI coding assistant, and its main job is to spend fewer tokens. It drops tool definitions you are not using and compresses large JSON tool outputs, which is where most of the savings show up. Repeated queries come back from a semantic cache in under 300ms, so they cost nothing. Routing is by task difficulty: simple prompts go to a free local model like Ollama, while harder ones escalate to your subscription or API, and a conversation stays on one model until the work genuinely outgrows it. The wrap mode is what makes this practical for Claude Code, Cursor, and Codex users, since it needs no code changes, just a base URL override or a single command. If you pay per token or keep hitting subscription limits on agentic workloads, this is a straightforward way to stretch that budget without replacing the tools you already use.