Reasonix: a terminal coding assistant built to stay cheap
Reasonix
Your AI coding assistant gets pricier the longer you leave it running. Reasonix is a coding assistant for your terminal, built so a long, leave-it-open session stays cheap instead. The whole design comes down to one habit: never changing the opening of what it sends the model.
A coding assistant that lives in your terminal
A coding assistant that lives in your terminal.
Here’s Reasonix in action. It’s an AI coding assistant: you chat with it, and it writes and fixes real code for you. You run it right in the terminal, the plain text window developers work in. There’s a desktop app and a code-editor plug-in too, and underneath they’re all one engine on your own machine. All of it built around one idea about cost.
An AI model forgets everything between messages
An AI model forgets everything between messages.
An AI model has no memory between messages. Every time you hit send, the model has forgotten the whole conversation so far. So to keep the model current, the assistant quietly re-sends the entire chat every turn: everything you’ve typed, and everything the model has said back. Early on, that’s cheap. But an hour in, each new message re-pays for that same growing pile of history.
The model re-reads the opening for about a tenth of the price
The model re-reads the opening for about a tenth of the price.
Here’s the catch, and the clever part. Reasonix is tuned for a model made by a company called DeepSeek. DeepSeek keeps a copy of the opening of each request, and re-reading from that copy costs about a tenth of reading it fresh. That saved copy has a name, the cache, and DeepSeek’s own pricing makes it roughly ten times cheaper. There’s one condition: the opening has to match the last one exactly, character for character. And that’s where most coding tools throw the discount away. They rewrite their own opening every turn: a fresh timestamp, the tool list reshuffled, the machine scanned again. The model sees a changed opening, misses the cache, and pays full price.
Reasonix keeps that opening perfectly still
Reasonix keeps that opening perfectly still.
trimmed
Reasonix holds that opening completely still, and it does it three ways. First, at startup it writes a short note about your computer, which tools you have, which shell, and pins that note at the very front. It saves the note and reuses it, so a restart can’t quietly rewrite it. Second, the list of actions it’s allowed to take stays in one fixed order, so that never shifts either. Third, when the chat gets long, it trims from the end, never the front. It tucks the older output behind a placeholder and reopens it only if it’s needed. All three keep the opening identical, so the discount keeps paying, message after message.
It measures its own cache hits and cost per task
It measures its own cache hits and cost per task.
Reasonix measures the payoff, and it shows you. It runs a set of test tasks and tracks how often the cache hits and what each task costs. The desktop app shows that hit rate live, as you work. Here’s one session: it reused that saved opening eighty-seven percent of the time, so most of the conversation cost the cheap tenth instead of full price. So the number is right there on your screen, adding up as you go.
Three ways to run it, and not tied to one model
Three ways to run it, and not tied to one model.
That desktop app is only one way in. There’s the terminal, the desktop app, and a plug-in that lives inside your code editor, all driven by the one program on your machine. And Reasonix works with other AI models too, not only DeepSeek. The savings just need a model that offers this same discount for re-reading the opening.
A single program for Mac, Linux, and Windows
A single program for Mac, Linux, and Windows.
Getting it running is quick: one command, and you’re in. It’s one self-contained program, with nothing else to set up. It’s free and open under the MIT license, so you can use it, change it, and build on it. The latest version shipped today, and new ones are landing fast.
Keep the opening still, and a long session stays cheap
Most AI assistants get pricier the longer they run. Reasonix keeps a long session cheap.
So that’s Reasonix. Most AI assistants get pricier the longer you run them. Reasonix keeps the opening of every request identical, so the discount holds and a long session stays cheap. The link’s on screen. New repo tomorrow.
