#Lynkr — open-source LLM router for every AI coding tool

1 messages · Page 1 of 1 (latest)

median nova
#

If you're using Claude Code, Cursor, Codex CLI, OpenClaw, Cline, Continue.dev, Windsurf, or Copilot — you're probably paying full price for every task. A typo fix costs the same as a full system refactor. All hitting the same expensive model.

Lynkr fixes this. It sits between your tool and any model provider. Classifies every request by complexity and routes automatically:

🔹 Simple tasks (rename, format, lint) → fast, cheap or local model
🔹 Medium tasks (tests, debugging, explanations) → mid-tier model
🔹 Complex tasks (refactor, architecture, multi-file) → most capable model

You change one config line. Your tools work exactly the same. Lynkr handles everything underneath.

#

Providers (12+):
Local → Ollama, llama.cpp, LM Studio, MLX Server (Apple Silicon)
Cloud → AWS Bedrock, OpenRouter, Azure, OpenAI, Vertex AI, Databricks, z.ai

Mix local + cloud in the same setup. Route sensitive code locally, heavy reasoning to the cloud.

#

What you get:
→ 60-80% cost savings — Claude Code users going from $150/mo to $30-50/mo
→ Semantic caching — similar prompts served instantly from cache, zero token cost
→ Long-term memory — context persists across sessions, fewer repeated calls
→ Fully local execution — no data leaves your machine, no ToS risk, no training data concerns
→ Circuit breakers + auto-failover — provider goes down, Lynkr switches automatically
→ Prometheus metrics — see exactly what's routed where and what it costs