vinhnx

For the past few weeks, I have spent my hobby as building and improving my own coding harness, VT Code. I plan to keep improving it throughout this month. Recently, I built a native macOS using GPUI (gpui.rs). Docs: https://github.com/vinhnx/VTCode/blob/main/docs/development/.... I'm pretty happy with how it's turning out, though there's still plenty to improve

I've been building my own open-source coding agent, VT Code, in Rust.

I wanted more control over the harness itself, especially tool execution, sandboxing, approvals, and context management.

Recently built a native macOS app with GPUI on top of the same agent runloop. It's been interesting exploring how the same engine works across terminal and desktop workflows.

https://github.com/vinhnx/VTCode

I've been building VT Code, an open-source coding agent written in Rust.

It focuses on structural code editing, sandboxed execution, and keeping developers in control of agent actions. Supports multiple LLM providers, with a CLI/TUI and a new native macOS app built with GPUI, both sharing the same agent engine.

Would love your feedback!

https://github.com/vinhnx/VTCode

I built a ChatGPT Site [0] for VT Code [1]: a browser-based WebMCP editor for exploring code, reviewing changes, and working with the agent.

[0] https://vtcode.vinhnx.chatgpt.site

[1] https://github.com/vinhnx/VTCode

I built VT Code in Rust with a similar goal: lightweight, terminal-first, and shipped as a single binary.

https://github.com/vinhnx/vtcode

I built VT Code in Rust for similar reasons. Single binary, low runtime overhead, with extensibility mostly handled through MCP.

https://github.com/vinhnx/vtcode

Been trying Claude Sonnet 5.5 through Merge Gateway. So far it feels fast and efficient, and noticeably different from Sonnet 5.

The communication and writing style also feels closer to Opus 5.5.

Hi Matt, awesome you're here. I've been using GLM-5.3 Flash since it came out and really enjoy using with via Merge AI Gateway.

I've been building VT Code for about a year as a way to understand how coding-agent harnesses work in practice. This month I'm continuing building it https://github.com/vinhnx/VTCode.

I've been using Merge AI Gateway and it's been useful so far. They tend to add new models quickly, and support has been responsive. https://gateway.merge.dev/

GPT-6 Astra scores 74.1% at DeepSWE v1.1 bench. Huge!

https://chatgpt.com/ is up again

It seems OpenAI is going to release Astra soon, https://x.com/ChatGPT/status/2095527989077557738.

I've been building VT Code for about a year as a way to understand how coding-agent harnesses work in practice.

Most of the code has been built with the help of coding agents, including VT Code itself. My focus has been on designing and iterating on the harness: tool execution, context management, sandboxing, permissions, verification, long-running sessions, subagents, and provider support.

It's written in Rust and runs primarily in the terminal.

https://github.com/vinhnx/VTCode

I'd be interested in feedback from others building coding agents, especially around long-running reliability and harness design.

Muse Spark 1.3 Max is the first Meta model to surpass OpenAI’s best on Artificial Analysis’ Intelligence Index. https://artificialanalysis.ai/models#intelligence

Worth noting: Claude Fable 5.1 and Mythos 5.1 are Anthropic’s first models to watermark text outputs.

I keep up with HN by reading it daily. My Firefox browser always have 2 pinned tabs at the top most: the front page /news and /ask. I also subscribed to HN Weekly Newsletter (https://buttondown.com/hacker-newsletter/archive/).

I have been building and maintaining a coding harness, with help from the community https://github.com/vinhnx/VTCode. Hope you'll check it out.

I think the paper is really worth reading for anyone working in software. As far as understand it is a software architecture paradigm where everything is a "plugin", and as plugin, I can plug-it-in and plug-it-out, if I understand it correctly. They called it "revertible effects", where software can strip out live code without a restart or system reboot. They give the example of VS Code, which requires system restarts whenever an extension or plugin needs to be updated. To help myself understand it, I created a quick video, using NotebookLM: https://www.youtube.com/shorts/LtR7DRlZJ0M.

The Cordis plugin architecture is interesting

https://github.com/cordiverse/paper

Learn by doing. Even more true nowadays, I think the best way to learn and understand how LLM and advanced topic like agent harness works is by, get your hands dirty and start building one. Trust me, it will pay off, one day.

I have been building a coding agent, VT Code, in my spare time. It's a way of me to self-research and explore how coding agent works in practice by building one. https://github.com/vinhnx/VTCode

This month marks 1 year since I first started self-research and building my own coding agent VT Code [0]. I've been learning and experiment with a lot of things to first adapt myself and to learn about AI native/harness engineering. Thankfully, last month VT Code receives highest amount of contributions from the open-source community.

This month I will be focusing and enhance more long-running horizontal for VT Code such as plan mode, better strengthen tools policy structure and improve its harness as new and existing model grows exponentially. And If you are curious and want to fork it feel free, and welcome, let's build!

[0] https://github.com/vinhnx/VTCode

Small model-facing surface, more responsibility in the harness. I'm exploring that tradeoff in Rust with VT Code: https://github.com/vinhnx/VTCode

I'm building roughly this in Rust: single binary, XDG-native, with Landlock/seccomp, scoped filesystem access, default-deny egress, and policy-based approvals. https://github.com/vinhnx/VTCode

Because Luna feels right and just enough for me.

This month marks 1 year since I first started self-research and building my own coding agent VT Code [0]. I've been learning and experiment with a lot of things to first adapt myself and to learn about AI native/harness engineering. Thankfully, last month VT Code receives highest amount of contributions from the open-source community.

This month I will be focusing and enhance more long-running horizontal for VT Code such as plan mode, better strengthen tools policy structure and improve its harness as new and existing model grows exponentially. And If you are curious and want to fork it feel free, and welcome, let's build!

[0] https://github.com/vinhnx/VTCode

GPT-5.6 Luna xhigh/max from Codex is my daily driver now. Good economics, good performance. It sits in the most attractive quadrant on the Intelligence Index.

I reached a mid-run usage cap and burn all my Plus subscription usage. I'm hoping there will be another reset.

Kimi K3's full weight is now live on Hugging Face https://huggingface.co/moonshotai/Kimi-K3