wrhn

Recent submissions

Comment · 2 months ago
Everyone agrees you should use the best model. But the current infrastructure makes this impossible. Look at what actually exists: the leading harnesses offer a handful of models. Open-weight and non-frontier models sit unused. Attempts to aggregate models discount complexity and fail to accommodate user diversity. The solutions come in two forms, and both are incomplete The first asks you to move to a new app, a new workspace, or ecosystem. But the best developers we know change tools constantly, because the best place to work keeps changing. Any solution that requires relocation is betting against how engineers actually behave. The second comes with a value tax, either charging for using your API keys, taxing and controlling how you use the product, or making assumptions around how you should use models that interfere with flexible deployment. Things that were frustrating to us were down-routing to models we hadn’t requested when they weren’t available. Making routing claims that don’t hold, getting detailed caching reasoning and traces, or controlling how we can use the tools, hooks, and mcps that we want to in the places we wanted to work from. We fixed it by building a truly universal pipe Cloud models, self-hosted models, custom models, your own keys, the hardware under your desk: it all connects the same way, and bringing your own costs nothing. We figured out how to bring models to the apps you already operate in without changing the setup and assumptions around your work. We’re really excited to build this with you and want to make everything we can open source and maintained by the community.

https://github.com/ConiferKit/use-conifer

Comment · 2 months ago
Are there any speed gains being drawn from different infrastructure providers or purely software
Comment · 5 months ago
It's a local inference engine we built from scratch in Rust - not another UI wrapper. The engine, the agent, and the tooling are all ours, tuned for your machine. This allows us to enforce security and data permissions instead of trusting a model's guardrails. One dependency keeps things simple and you can confidently source blame onto us if something breaks.

What's inside - Ecosystem not just a chatbot: Editor, integrated terminal, file viewers, IDE - the surfaces you actually build in. - Security: All agents and tool calls run inside a kernel-level sandbox, so it can only touch what you allow. This isn't something a prompt or agent controls, its a boundary the OS holds for you. - One click: We kept it simple, no docker, no servers to point at, no config files. Download the app, open it and its running. - Nothing leaves your computer: No telemetry, no accounts, no cloud latency, no network downtime.

thanks for being early. we expect things to break here but we're going to work everyday to fix it send us an email or fill out the website form for feedback: [email protected]

Comment · 5 months ago
Not hating because his project is very cool but this is just a python UI on top of other oss - I'm only saying this bc team at Princeton and I have been building wthis but ground up in rust with more utility and running more custom tools, app integration, custom kernels outperforming llamacpp much more here www.conifer.build

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton