jade panther
Lynkr is an open-source proxy that sits between your AI coding tools (Claude Code, OpenClaw, Cursor, Continue.dev, Codex) and any LLM provider. It scores every request by complexity, routes it to the cheapest model that can handle it, and compresses
tokens before they leave your machine.
One endpoint. 12+ providers. 80-95% cost reduction.
The Problem
Every request to Claude Code or Cursor costs the same — whether you're asking it to ls a directory or architect a distributed system. You're paying Opus/Sonnet prices for tasks a free local model handles perfectly.
How Lynkr Fixes It
Intelligent 4-Tier Routing
Every request gets scored on 15 dimensions (token count, technical depth, tool chaining, reasoning complexity) and routed to the right tier:
Score 0-25 SIMPLE → Ollama ,llama.cpp (FREE) "list files in src/"
Score 26-50 MEDIUM → GPT-OSS ($) "explain this function"
Score 51-75 COMPLEX → Gpt 5($$) "refactor auth module"
Score 76-100 REASONING → Kimi K2.5($$$) "design caching layer"
Most coding sessions are 70%+ simple/medium tasks. That's 70% of your bill eliminated.
6-Phase Token Optimization (60-80% reduction)
Before any request hits a provider, Lynkr runs it through:
Smart Tool Selection
Prompt Caching
Memory Deduplication
Response Truncation
Dynamic System Prompts
Conversation Compression
GitHub: https://github.com/Fast-Editor/Lynkr
red veldt
short fern