What is PennyRouter?

I initially built PennyRouter so I could get more Fable time. When Anthropic released their latest generation model, I was blown away by what it could do — and by how fast it burned through my subscription. So I started working on various nips and tucks to Claude Code to get more mileage out of it. I explored a wide range of plugins, wrappers, and skills, going way deeper into the world of token optimization and cache economics than I had ever intended. When I finally came up for air, I had crystallized on three core principles:

1. Cache warmth is by far the biggest factor in AI costs, and most people aren't even really aware of it.
2. There are dozens of small but beneficial tweaks that can improve token utilization, but you've got to mix them carefully to avoid derailing quality.
3. The average person, even a vibe-coding enthusiast, does not want to spend their time wrangling multiple repos that start with "spin up docker" or "set up a local proxy."

So I built PennyRouter for myself, creating a simple layer that pulled together the strategies I found worked best after testing hundreds of my own conversations and coding sessions. My goal was to get the same experience and the same code quality, just longer sessions and — when I used the API directly — lower costs. My steady state was savings of about 30% on short sessions and 60% on longer ones, meaning with my mix of work I could get about double the AI for my subscription compared to before.

Since I was mucking around, I also added a few features I just wanted to have while coding:

• The ability to import a conversation from ChatGPT, Claude, Gemini, Perplexity, etc. directly into a coding session (@penny import)
• Being able to consult a more advanced model without switching my whole session over (@penny consult)
• Being able to send a question to a colleague from within Claude Code so they could weigh in on a particular part of that session (@penny send message)
• The ability to transfer a session between Claude Code, Codex, Cursor, etc. so I could continue in any harness if my subscription ran out (@penny transfer session)
• A session warmth tracker that shows when your session is about to turn cold, so you can take action before the cache expires

I offered PennyRouter to a few friends to try out (AI harnesses is the kind of thing that comes up when you are a Georgia Tech professor). They had some suggestions, found some bugs, and very much helped me improve it to the point where I felt like I could share it more broadly. So that's what PennyRouter is now. I tried to keep the cost low but enough to cover the servers and such so I could sustain keeping it available.

I hope you'll give it a try and let me know what you think. If it saves tokens and gives you a few extra powers along the way, would love to hear it. And if it is doing weird things or slowing you down or not saving tokens, please let me know that too and I'll refund the $5 subscription fee. Contact me at [email protected].