@siah

UC Berkeley alum. Silicon Valley AI/ML Scientist. Formerly at Microsoft. Spent a decade building teams and products at Apple. Now solo

 US
Joined April 2007
  • Tweets210
  • Following1.3K
  • Followers7.6K
89Media
💻I have been working on a free and open source self-improving coding agent called Chorus. I built it for myself but I think it can be useful to others too. It provides scaffolding around Claude Code that allows you to deploy on a remote node, run loops, define specialized agents, manage tickets, etc. What I love about it is the self healing of agents where another agent can read the agent logs and improve them. It is free and open source. I also have a video demo that explains the features youtube.com/watch?v=XtsupeZI…
2
1
1
9
1,276
AI needs its GNU moment: a full stack of tools that are free and open all the way down. Open models bundled with any Linux distro, open-source coding tools to replace Claude Code and Codex, and open agents and computer-use models that anyone can run, inspect, and modify.
3
482
I am constantly heartbroken that there is no emoji for a “funnel” 🔽
1
308
This is so good and the fact that we can create these locally on a consumer GPU is unbelievable
cocktail peanut
@cocktailpeanutSep 8
1
751
I’ve been really enjoying connecting Codex and Claude Code to our Hue lamps! But once again, I walked into our bedroom at 3 a.m. last night and realized our lamps had been pulsating the whole night while I was working in our home office. Thankfully my wife sleeps with an eye mask! Our cat was pretty confused, though.
Mark Faridani
@siahJul 14
Added a ton of new behavior to Agent-Glow 💡 I personally find Agent Glow to be pretty useful for when I use Claude Code or Codex. We have a number of smart lights at home, and I can leave a coding session running on my computer at the office, and as soon as the AI coding agent is done, the lights turn green. If the agent is waiting for my confirmation, the lights turn red. And while the agent is thinking, the lights pulsate like slow breathing. I really love it and have been correcting a lot of its behavior. It supports both Philips Hue and WiZ protocols. I hope you also find it useful
1
2
1,267
Does anyone here have a #Zoox invite? My wife really loves to try it
355
Added a ton of new behavior to Agent-Glow 💡 I personally find Agent Glow to be pretty useful for when I use Claude Code or Codex. We have a number of smart lights at home, and I can leave a coding session running on my computer at the office, and as soon as the AI coding agent is done, the lights turn green. If the agent is waiting for my confirmation, the lights turn red. And while the agent is thinking, the lights pulsate like slow breathing. I really love it and have been correcting a lot of its behavior. It supports both Philips Hue and WiZ protocols. I hope you also find it useful
Mark Faridani
@siahJul 4
I created a hook for Claude Code and Codex that makes our Hue bulbs pulsate when the AI coding agent is thinking. The lower bulb turns red when Claude is waiting for my response. Pretty useful for those who want an inch of work-life balance.
1
4
2,125
Made a little parametric design tool for creating those modern looking vases #fable5 #3dprinter
1
1
8
1,633
try it here faridani.github.io/turny/
1
168
I created a hook for Claude Code and Codex that makes our Hue bulbs pulsate when the AI coding agent is thinking. The lower bulb turns red when Claude is waiting for my response. Pretty useful for those who want an inch of work-life balance.
4
2
2
16
1,919
The repository is here: github.com/faridani/agent-gl…

GitHub - faridani/agent-glow: Hue meets your coding agent

Hue meets your coding agent. Contribute to faridani/agent-glow development by creating an account on GitHub.

github.com
142
The repository is here on github and is Apache 2 github.com/faridani/chorus

GitHub - faridani/chorus: Chorus is a self-hosted, 24/7 AI engineering team. A single orchestrator...

Chorus is a self-hosted, 24/7 AI engineering team. A single orchestrator coordinates role-based agents—dev, QA, designer, and more—each with defined guardrails. Powered by subscription CLI tools li...

github.com
1
109
I've been heavily stress testing file download stability in Firefox vs Chrome (simply because I still naively believe Computer-Use is the last puzzle AI needs to solve for us and it hasn't). Firefox is genuinely one of the most resilient downloaders out there. It is unbelievable.
1
3
488
Too bad firefox is usually identified as robot by websites :(
220
Most of the newly released Gemma 4 models are small enough to run on a strong mac or an RTX gaming card ollama.com/library/gemma4/ta…

Tags · gemma4

Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.

ollama.com
1
345
I have been battling latency challenges with AI/LLMs. Anyone who has worked on putting these into real-time systems knows you must navigate the latency-accuracy tradeoff. It is a huge challenge and hard to navigate. Frankly, we don't have any solution for it other than dumbing down the LLM by using smaller or distilled models.
1
2
553
Claude Code is down!
812
This is now getting exciting
Sunny Patneedi
@sunnypatneediFeb 23
Replying to @karpathy
@karpathy fit GPT training in 260 lines with microgpt. Fit a full coding agent in 407 with help from @siah and @claudeai microcodeagent is one Python file, zero dependencies, 7 tools, complete agent loop. 7 tools: think, bash, read, write, edit, find, and grep. The entire algorithm is a while loop. Run python3 microcodeagent.py on any machine with Python 3.8+ and an API key. It works. No install step. gist.github.com/sunnypatneed…
1
748
Is there any LLM benchmark that also reports `Time To Last Token (TTLT)` ? I think an end-to-end latency metric can help replace a lot of old school ML with LLMs
388
More than 30,000 people have been killed by the Islamist regime in Iran. You can read their stories and be their voice #FreeIran 💚🤍♥️✊🏾 instagram.com/p/DUEO-SXga1I/…
1
1
379
Load more