khazra

Front-end engineer

[khazra @UnderTheGun5]

@0x_ultra Do they at least have a privacy policy where all of this is disclosed? This would be fine-ish if people would have to opt-in. If it's the default and hidden from users, this is a major privacy violation.

[Ahmad @TheAhmadOsman]

DROP EVERYTHING

The bible for running LLMs locally is now available online to read for free

Covers what to use on

- Laptop / edge / odd hardware
- Mac-first workflows
- Single RTX GPUs
- 2-4+ NVIDIA / CUDA GPUs
- General production serving
- Long-context / MoE / routing
- NVIDIA max performance
- Cluster orchestration

Software

- llama.cpp
- MLX / MLX-LM
- ExLlamaV2
- ExLlamaV3
- vLLM
- SGLang
- TensorRT-LLM
- NVIDIA Dynamo

You should read this, and if you cannot now then you most definitely wanna bookmark it for later

Local AI FTW

[fucory @FUCORY]

1/ Lots of people have asked me what I think about Claude’s new workflows.

My take: they’re excellent.

I’m already using them.

I’m so energized to see the thesis behind Smithers getting validated.

[TRAE @Trae_ai]

https://t.co/vdDiOj8k7l

[fucory @FUCORY]

🚀 Smithers 0.14.0 just dropped

New features. No more crashed processes. No more exploding DBs. No more “why is it stuck?”

🧵

[khazra @UnderTheGun5]

@sqs @p0 Hey Quinn, `amp -m internal` doesn't work for me, I want to be fast and internal as well!

[fucory @FUCORY]

It sounds so gimmicky

AI Orchestration framework via React

But hear me out, Smithers is secretly the best-in-class agentic orchestration framework

In this 🧵 I answer two questions

1. Why you even need an orchestration framework?
2. Why that orchestration framework should be smithers

🧵 1/18

[George @odysseus0z]

https://t.co/88l0mFNMIE

[khazra @UnderTheGun5]

Holly... WTF

[The White House @WhiteHouse]

STRIKE. 💥🦅

[The White House @WhiteHouse]

STRIKE. 💥🦅

[wevm @wevm_dev]

Introducing 𝚒𝚗𝚌𝚞𝚛 – the CLI framework built for agents and humans.

Automatic discovery for agents enabling a guided experience for humans, without compromising tokens & context windows.

» npx incur skills add

[Nikita Bier @nikitabier]

@cahimalyanjoshi In the last 5 months since joining X, you have posted 39,800. That is 245 posts per day.

Assuming you sleep 8 hours, that is one post every 4 minutes since registration—with no breaks.

If spam was an Olympic event, I have no doubt that you would be a Gold Medalist.

[Sahil Bloom @SahilBloom]

https://t.co/3RufL4ztsU

[Connor Boyack 📚 @cboyack]

https://t.co/4SMw7birVU

[fucory @FUCORY]

Now proudly introducing, the first ever vibecoded Ethereum execution client

It's actually 2 clients
- Zig client
- @EffectTS_ client

If you want to see how the AI Slop Sausage is made, follow XVI

1/2

[Gordon Mickel @gmickel]

Looked at Anthropic's ralph-wiggum plugin again. Single session, accumulating context, no re-anchoring.

This really defeats the purpose.

The original vision by @GeoffreyHuntley:
- Fresh context per iteration
- File I/O as state (not transcript)
- Dumb bash loop, deterministic setup

Anthropic's version? Stop hook that blocks exit and re-feeds the prompt in the SAME session. Transcript grows. Context fills up. Failed attempts pollute future iterations.

The irony: Anthropic's own long-context guidance says "agents must re-anchor from sources of truth to prevent drift."

Their plugin doesn't re-anchor. At all.

---

flow-next follows the original vision:
✅ Fresh context per iteration (external bash loop)
✅ File I/O as state (.flow/ directory)
✅ Deterministic setup (same files loaded every iteration)

Plus Anthropic's own guidance:
✅ Re-anchor EVERY task - re-read epic spec, task spec, git state
✅ Re-anchor after context compaction too (compaction shouldn't happen but if it does, we're set)

Plus what we added:
✅ Multi-layered quality gates: tests, lints, acceptance criteria, AND cross-model review via @RepoPrompt
✅ Reviews block until SHIP - not "flag and continue"
✅ Explicit plan → work phases - plan reviewed before code starts
✅ Auto-blocks stuck tasks after N failures
✅ Structured task management - dependencies, status, evidence

Two models > one. Process failures, not model failures. Agents that actually finish what they start.

https://t.co/4JXXXYg8eO

[khazra @UnderTheGun5]

@elie2222 @cursor_ai @claude_code @cubic_dev_ @baz_scm @coderabbitai @detaildotdev @vltansky Thanks for this. It's very valuable to see a real-life, working SWE AI workflow. Do you always use the same context window (chat) in which the Cursor's plan is created when you execute it?

[Quinn Slack @sqs]

Opus 4.5 is free in Amp (up to $10/day)

https://t.co/bzCtHP6iV5

[fucory @FUCORY]

The most radical difference of Voltaire to older libraries is we do not include any high level abstractions

We instead provide tested reference implementations and propmts to allow you to generate your own abstractions specific to your codebase

[fucory @FUCORY]

Introducing ai-zig library

A zig port of the popular vercel ai-sdk library with support for 30+ AI providers