capocasa
Re:Nim Version 2.4.0 Released
on Wednesday October 28, 2026 @02:03PM (#50278152)
Oh my goodness! Excitement!!!
Re:Show HN: Open-source model routing for coding agents at Astra-level...
on Friday October 02, 2026 @05:06AM (#49931398)
That sounds really quite interesting- one question pops up, this sounds like it could produce a lot of non-cache input tokens. How much impact does this have, and are you mitigating it?
Interested in adding support to my own 3code- https://3code.capocasa.dev- which aims to reduce cost by using more compact system prompts and a clever compaction variant.
Thank you!!! And thank you for your interest!
Yes- I plan to make a a full web app, as well as a web-and-terminal orchestrator app. Still planning the details.
I'm here for the long haul!
I believe software can be done- with incremental improvement and keeping the foundation, like vim. That's the 3code trajectory.
New fun ideas go into seperate tools, possibly ones 3code (or other minimal agents) can call, but the core stays lean.
I agree! MCP should be a command line tool wrapper than can be used from any minimal agent without bloating context if you don't need it.
I did some token tests and both pi and my own 3code used 4x fewer tokens than Claude Code, while opencode used 2x fewer tokens than Claude Code. You can tell Anthropic just doesn't care about using tokens and the tradeoff is maximal performance. That's understandable, but it might not be the tradeoff you need, hence, alternative clients.
I went for a middle ground in https://3code.capocasa.dev.
Each family gets their own system prompt, based on the original harness one, but compacted and includes instructions for more brevity.
I made just that!
Written in Nim. Unixy scrollback but cross plaform. Tested for token efficiency.
Or for something not quite as bare bones- batteries included and tested for token consumption- https://3code.capocasa.dev
This Year I Vibecoded My Taxes
Re:Mistral raises €3B
on Tuesday September 08, 2026 @07:54AM (#49609067)
Last time I tried Mistral large it was pretty bad at programming but I hear they're fairly good at European language legal.
10-task GLM 5.3 harness bench: Claude, OpenCode, pi, zcode, Hermes and 3code
Re:Private German rocket makes history, reaches orbit from European soil
on Saturday September 05, 2026 @05:23PM (#49580821)
This is such a breath of fresh air
Re:Show HN: FrontierHarness Eval – 9 harness, same model, cost per pas...
on Thursday September 03, 2026 @09:15AM (#49549548)
I saw that very much developing 3code. Even slight system prompt variations can have huge differences in results, so I have one system prompt per model family.
3code benches 75 lower token use vs. OpenCode on 10-task SWE-Bench subset
Re:Are we offloading too much of our thinking to AI?
on Tuesday July 14, 2026 @05:15PM (#48913042)
Definitely, but since my experience is that without the AI the thinking simply wouldn't happen at all in most cases it's a net improvement.