Edgee
Take back control of your token spend
Start new thread

Edgee Mod for Claude Code - Use any model inside Claude Code, keep your Claude plan

by
•
A Claude Code mod that puts Edgee in your terminal. Keep your Claude plan, switch your session to any of 200+ models in one keystroke, and see live what compression and routing save you. Same conversation, same context. Two commands to install.

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt 👋 Sacha here, CEO of Edgee. Like a lot of you, we spend our days in Claude Code, on a Claude plan. It's great. But some tasks deserve another model: a cheaper one for routine work, a specialized coder, or the new model everyone is talking about this week. Until now, trying one meant leaving your session and losing your context. So we built the Edgee Mod for Claude Code. It uses Claude Code's new mods system, so it runs inside Claude Code itself: 🔀 Switch models in one keystroke. Open the Edgee panel, search 200+ models, press Enter. The next request goes to that model, in the same conversation. One click on "Claude Code's choice" and you're back on your plan. 💸 See what you save. A live cost savings donut splits what compression and routing save you, per session. 📊 Know what's happening. Requests, tokens, cache hits and models served, right next to your conversation. Folded, it's a single line above your prompt. Getting started takes two commands: curl -fsSL | bash edgee launch claude Fun timing: Mistral Large 4 just came out and it's already available in Edgee. It's the model we switch to in the demo video. We'd love your feedback: which model would you try first inside Claude Code? I'll be here all day to answer questions.
Congrats on your launch! 🚀
Maker
thanks a lot

congrats on the launch!

different angle from the compression thread, two quick ones.

requests go through your gateway so whats retained? i work on a product where customer data cant leave the tenant, and someone always has to answer 'where did the code go' before we adopt anything.. self host, or logging you can turn off??

also, when a request routes to another model mid conversation there's no warm prefix on that side, so that call should be expensive. does the savings donut account for it?

(btw i'm not AI 😂)

switching models mid-conversation and keeping "same context" is the part I'd want to pressure-test. compression is probably fine for prose, but code context is unforgiving - a dropped variable name or a compressed-away line of a diff doesn't just lose nuance, it can make the next model's edit wrong in a way that's hard to notice until it breaks. is the compression ratio different for code blocks versus natural language in the conversation, or does it treat the whole context the same way?

Maker
Thanks Gal, that’s a fair concern. Our compression layers don’t touch code at all, so we don’t apply the same compression across the whole context. Code is left unchanged, rather than compressed at a different ratio.

 good to hear code is excluded entirely rather than just compressed less. one follow up - how do you detect what counts as "code" in a mixed message, like a prompt that pastes a stack trace or a config file inline with prose explaining it. is that a simple fenced-block detector, or does it have to be smarter than that to avoid compressing something that's technically not in a code block but still breaks if touched.