it is fast, there is no fucking gateway fuckaround or any other similar issues, up and going in seconds
straight away I added two skills, that it wrote for itself, read my gmails and attachments, and browse the web, text browser first up, render page and screenshot with OCR for javascript heavy pages\
then I asked it, find the best value ram in my area, second hand as well as new, try gumtree and facebook marketplace, plus anything else relevant, bam - 15 seconds maybe a concise summarised range of options
then on another project, I told it to /study and then used the gmail plugin to access all the relevant gmails and attachments (which included minutes of all the meetings) and it was fully up to date with the project I am working on and ready to go
best agent I have used so far by a country mile, if you don't try it then that is your loss
did I mention it was fast, like 3x to 5x better productivity fast compared to openclaw, at least
one thing it does not do is support the up arrow/down arrow to scroll thru past commands, but you can just tell it, "run that websearch for ram again" etc, i will totally live witht his for all the other positives
Zot seems interesting, this is the first time I see it. On a quickl look it seems like Pi, but in Go. I was hoping to embed Pi into some of our internal projects and the typescript stuff was blocking me, I'll definitely give Zot a look.
pi is awesome, quite possibly the best OSS tool out there. You should definitely give it a shot if it fits your stack. zot has become my daily driver. I didnt build zot to compete. I built it to really get a feel for how harnesses work, and I do it with Go simply because I love the language. More on that here: https://www.patriceckhart.com/blog/posts/2026-04-23/why-i-bu...
I'm getting a little fatigued by all the harnesses that are made by other coding agents. Like, when I checked out opencode, it looked and felt incredibly impressive, until I looked at how frequently it completely invalidated the KV-cache. After looking at the source code, it's basically unsalvageable and I ran far far away. (It's mostly imperative garbage which is typical of undisciplined agent output. It doesn't even use React, it uses some other reactive library in a non-declarative way, I think SolidJS)
DeepSeek Reasonix is better in terms of cache stability because that is a core tenet, which should honestly be table stakes for agentic tooling, but the TUI is kind of ugly and the tools also kind of suck (they pretend the sandboxed working directory is at /, which makes the model almost unable to use MCP servers that expect to be passed filesystem paths). On top of that, it doesn't expose the structuredContent of MCP server tool responses, which is like... the entire point of it? Now all my tools that return huge swaths of JSON data into structuredContent, which Claude Code can process perfectly fine, need an additional separate path to generate readable versions of it into content because Reasonix ignores structuredContent for some reason. That's supposed to be the model-side output, while content is the user-side output, but whatever.
I don't know how much more of this I can take. I'm in the process of working on my own harness essentially from scratch, manually, because I'm so fed up with all this vibecoded tooling that misses incredibly basic and obvious design.
I feel like Claude Code used to be from scratch like this and that was why it was so good, until they started vibecoding large swaths of it and stripping away all the power-user features and good taste that made it so wonderful before. Now it even has random, inexplicable problems like "API Error: 400 messages.1.content.15: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response." which shouldn't even be able to happen!!
And like, I get the distillation angle of why thinking output was completely removed from Claude, but I work in bypass-permissions mode and I want to correct misunderstandings as I see them. This is different than wanting to review each edit.
Speaking of reviewing each edit, I hate that Reasonix doesn't print diffs, and just says "use git diff". Like, no? I want to see each change the agent made and when. I don't want to only see one diff at the end; that nearly ruins the point of conversation history.
I've deliberately been post-poning harness building.
I think it's great as an obligatory learning experience.
But I'm hoping someone will come along and provide the "best of breed" harness:
- OpenCode's TUI and client-server model,
- Claude's prompt engine,
- Pi's extensibility, and
- the codebase stability of a craftsman (yet to be seen).
I haven't tried other harnesses than those three. It's time-consuming, and does not align with my primary goals.
I've been reimplementing a TUI library based on Ratatui, but drawing the UI components of OpenCode's OpenTUI and a bunch of Ratatui-adjacent components. Was hoping someone would separate the concerns and reverse engineer Claude's prompt engine and just not provide a UI for it. Make it modular so each part can be replaced by something better. There's only really 3 parts: TUI library, engine, and client-server (so you can choose between web or terminal, and so you can host the engine + server in the cloud, resume your sessions, and whatever enterprise features you want for session and memory management.
Glad to see tooling in my native language. I don’t want to touch TypeScript stuff with a ten foot pole, but sadly it seems to be the lingua franca for agentic tools.
The one thing that would keep me from making the jump is CC’s auto mode.
zot is a coding agent harness. not a data vault, not a pacemaker, and not a life-support device in any medical sense.
Ive been coding for almost 20 years, and for the past few with Go. Nobody would believe that a project of this scale or even a much smaller one could be pulled off, halfway stable, over a couple of days. Not even with a blueprint or two in hand. Thats why it matters, and its totally fair, to point out when something is largely vibe-coded. "Vibe slopped" is meant more as a joke. The essential parts of the code I actually understand. Some of them I modified and overhauled myself.
zot is a learning project not production logic with peoples sensible data or lives depending on it. ;-)
Since this project hasn't had much attention, I replaced the submitted title ("Zot now supports Claude Opus 4.8") with that of https://news.ycombinator.com/item?id=47931161. I hope that's ok!
This real world usage of LLM's favorite word shows why LLMs pair this word with "what I'm about to say is seriously unreliable".
Here, "honest description" means the author didn't make something out to be more than it is. Perfect.
Ironically, LLMs don't apply that to the thing they're describing, but to their description itself. Meaning: when they say "honestly" it flags they have no idea and are about to be lazy, make it up, and confidently assert nonsense.
It's easiest to understand if you mentally insert a phrase:
"Honestly [you should disregard this because I am just making this up but], you made a great choice."
Coding agent harnesses strike me as similar to blog generators. They can be as simple or as complex as you’d like. Plugins help with adoption. And if you want it’s real easy to write your own that does exactly what you want.
Harnesses aren't really going to change much of the performance on models like Opus, and GPT.
You literally can just give the model a bash tool and it will do just fine in fact it will most likely do better than majority of harnesses due to how well models are at bash.
The model do all the lifting. It really doesn't matter which harness you use.
> Subscription-capable - Anthropic Claude Pro/Max (anthropic), OpenAI Codex / ChatGPT Plus/Pro (openai-codex), Kimi Code (kimi), and GitHub Copilot (github-copilot).
Am i reading this right? Seems to suggest that this can be used with Claude Code Subscription, which isn't true i think. Did this pre-date the CC Subscription change? Or is it playing fast and loose with the rules hah.
Maybe it's using `-p`, which technically works for another few days i think lol. (That's going away.. what, June 1st? Something like that?)
It does not use -p, but it does try to impersonate Claude when talking to the Anthropic API. Will they detect the difference in usage patterns and ban anyone who exploits them? Who knows.
I'm all for people writing their own coding agent harnesses... is there anything different about this one? Its not clear why I'd choose this over pi, opencode, or other existing options
Great minds think alike? Two months ago I created an agent called 'zop' [1] that's also a static Go app. It's not a code harness, it's a cli tool for quick one-liners (faster and less memory than opencode --prompt) with canned system messages. With compile tags you can strip it down to just prompt execution and the binary's less than 3MB.
....But also because feature creep, you can compile-in text-to-speech, speech-to-text, an interactive mode, an Android app, MCP/tool calling, multiple provider support, and now a really crappy web interface that only half works. It turns out vibe coding is harder/more time-consuming than it seems... Creating an alternative to beads made it more manageable, but I need multi-agent orchestration to code it so I don't have to babysit it and manually QA it (because just installing playwright and telling the AI to write tests doesn't really work).
Kind of a waste of time, but interesting learning experience. Now I know why there aren't a hundred magically awesome user tools out there... they're still not that easy to make.
37 comments
[ 3.6 ms ] story [ 79.0 ms ] threadit is fast, there is no fucking gateway fuckaround or any other similar issues, up and going in seconds
straight away I added two skills, that it wrote for itself, read my gmails and attachments, and browse the web, text browser first up, render page and screenshot with OCR for javascript heavy pages\
then I asked it, find the best value ram in my area, second hand as well as new, try gumtree and facebook marketplace, plus anything else relevant, bam - 15 seconds maybe a concise summarised range of options
then on another project, I told it to /study and then used the gmail plugin to access all the relevant gmails and attachments (which included minutes of all the meetings) and it was fully up to date with the project I am working on and ready to go
best agent I have used so far by a country mile, if you don't try it then that is your loss
did I mention it was fast, like 3x to 5x better productivity fast compared to openclaw, at least
one thing it does not do is support the up arrow/down arrow to scroll thru past commands, but you can just tell it, "run that websearch for ram again" etc, i will totally live witht his for all the other positives
DeepSeek Reasonix is better in terms of cache stability because that is a core tenet, which should honestly be table stakes for agentic tooling, but the TUI is kind of ugly and the tools also kind of suck (they pretend the sandboxed working directory is at /, which makes the model almost unable to use MCP servers that expect to be passed filesystem paths). On top of that, it doesn't expose the structuredContent of MCP server tool responses, which is like... the entire point of it? Now all my tools that return huge swaths of JSON data into structuredContent, which Claude Code can process perfectly fine, need an additional separate path to generate readable versions of it into content because Reasonix ignores structuredContent for some reason. That's supposed to be the model-side output, while content is the user-side output, but whatever.
I don't know how much more of this I can take. I'm in the process of working on my own harness essentially from scratch, manually, because I'm so fed up with all this vibecoded tooling that misses incredibly basic and obvious design.
I feel like Claude Code used to be from scratch like this and that was why it was so good, until they started vibecoding large swaths of it and stripping away all the power-user features and good taste that made it so wonderful before. Now it even has random, inexplicable problems like "API Error: 400 messages.1.content.15: `thinking` or `redacted_thinking` blocks in the latest assistant message cannot be modified. These blocks must remain as they were in the original response." which shouldn't even be able to happen!!
And like, I get the distillation angle of why thinking output was completely removed from Claude, but I work in bypass-permissions mode and I want to correct misunderstandings as I see them. This is different than wanting to review each edit.
Speaking of reviewing each edit, I hate that Reasonix doesn't print diffs, and just says "use git diff". Like, no? I want to see each change the agent made and when. I don't want to only see one diff at the end; that nearly ruins the point of conversation history.
I've deliberately been post-poning harness building.
I think it's great as an obligatory learning experience.
But I'm hoping someone will come along and provide the "best of breed" harness:
I haven't tried other harnesses than those three. It's time-consuming, and does not align with my primary goals.I've been reimplementing a TUI library based on Ratatui, but drawing the UI components of OpenCode's OpenTUI and a bunch of Ratatui-adjacent components. Was hoping someone would separate the concerns and reverse engineer Claude's prompt engine and just not provide a UI for it. Make it modular so each part can be replaced by something better. There's only really 3 parts: TUI library, engine, and client-server (so you can choose between web or terminal, and so you can host the engine + server in the cloud, resume your sessions, and whatever enterprise features you want for session and memory management.
The one thing that would keep me from making the jump is CC’s auto mode.
Ive been coding for almost 20 years, and for the past few with Go. Nobody would believe that a project of this scale or even a much smaller one could be pulled off, halfway stable, over a couple of days. Not even with a blueprint or two in hand. Thats why it matters, and its totally fair, to point out when something is largely vibe-coded. "Vibe slopped" is meant more as a joke. The essential parts of the code I actually understand. Some of them I modified and overhauled myself.
zot is a learning project not production logic with peoples sensible data or lives depending on it. ;-)
https://news.ycombinator.com/newsguidelines.html
(I also merged https://news.ycombinator.com/item?id=47941645 from that thread into this one)
Here, "honest description" means the author didn't make something out to be more than it is. Perfect.
Ironically, LLMs don't apply that to the thing they're describing, but to their description itself. Meaning: when they say "honestly" it flags they have no idea and are about to be lazy, make it up, and confidently assert nonsense.
It's easiest to understand if you mentally insert a phrase:
"Honestly [you should disregard this because I am just making this up but], you made a great choice."
You literally can just give the model a bash tool and it will do just fine in fact it will most likely do better than majority of harnesses due to how well models are at bash.
The model do all the lifting. It really doesn't matter which harness you use.
Am i reading this right? Seems to suggest that this can be used with Claude Code Subscription, which isn't true i think. Did this pre-date the CC Subscription change? Or is it playing fast and loose with the rules hah.
Maybe it's using `-p`, which technically works for another few days i think lol. (That's going away.. what, June 1st? Something like that?)
....But also because feature creep, you can compile-in text-to-speech, speech-to-text, an interactive mode, an Android app, MCP/tool calling, multiple provider support, and now a really crappy web interface that only half works. It turns out vibe coding is harder/more time-consuming than it seems... Creating an alternative to beads made it more manageable, but I need multi-agent orchestration to code it so I don't have to babysit it and manually QA it (because just installing playwright and telling the AI to write tests doesn't really work).
Kind of a waste of time, but interesting learning experience. Now I know why there aren't a hundred magically awesome user tools out there... they're still not that easy to make.
[1] https://codeberg.org/mutablecc/zop