# Lynkr

An LLM gateway and CLI proxy that compresses tool outputs, caches semantically similar requests, and routes each request across more than a dozen providers by complexity, including wrapping an existing Claude Code or Codex subscription.

Lynkr is a proxy that sits between an AI coding tool and its model provider to cut token usage and cost. It compresses large JSON tool results, caches semantically similar requests so a near-duplicate query can be answered from cache in under a second, and routes each request to a tier, such as a free local model, a cheaper API, or a paid subscription, based on how complex the task appears, keeping one routing decision per conversation with automatic escalation if a task turns out to need more capability.

Its wrap mode intercepts an existing coding assistant, such as Claude Code, Codex, Copilot, Aider, or Cursor, so that simple requests are routed to a local or lower-cost model while the user's own subscription or API budget is reserved for harder tasks, without changing how the assistant itself is used. An interactive setup wizard configures usage mode, tier assignments across the supported providers, credentials, and routing behavior.

It supports fourteen model providers, including local runners such as Ollama and llama.cpp, and is installed as a global npm package.

Website: http://lynkrhq.com/