Mario Zechner's banner
Mario Zechner's profile picture

Mario Zechner

@badlogicgames • 51,898 subscribers

Armin's handler at https://t.co/B05ybKGkzx. Old man yelling at Claudes. https://t.co/Q1wG57v1yc https://t.co/mnOoWUr0TO https://t.co/8i5vIRE0Wn

Shorts

flicker company taketh, giveth. sound on for euromaxxing vibes.

flicker company taketh, giveth. sound on for euromaxxing vibes.

28,719 views

had to use violence. but the legs will be a great base for an old Android phone.

had to use violence. but the legs will be a great base for an old Android phone.

22,596 views

felt cute, did some Andrej Karpathy style cozy coding. now i can run GPT 2 124M in pure TypeScript at 7 tps. played with implementing the GEMV via C/WASM, but that only got me a 1.7x speed up.

felt cute, did some Andrej Karpathy style cozy coding. now i can run GPT 2 124M in pure TypeScript at 7 tps. played with implementing the GEMV via C/WASM, but that only got me a 1.7x speed up.

29,165 views

sick and tired, so i can only vibe slop. feeling the agi in my non-crdt, non-ot collaborative markdown editor thing. both opus and gpt should have implemented this: both failed, hard :D

sick and tired, so i can only vibe slop. feeling the agi in my non-crdt, non-ot collaborative markdown editor thing. both opus and gpt should have implemented this: both failed, hard :D

35,582 views

Videos

LIVE
1.2k

Anya Rossi

sweetdream.ai

SweetDream.aiSponsored•Livecam
Streaming Now

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

HD live stream
Exclusive private shows
1.2k viewers online

Free preview available • Premium content

introducing pipi, the shitty robot. brain lives on my laptop, sensors/UI live on the mounted phone. time to completion: 24h (minus sleep, knight festival, lunch, dinner, and play) built with
3:12
badlogicgames's profile picture

introducing pipi, the shitty robot. brain lives on my laptop, sensors/UI live on the mounted phone. time to completion: 24h (minus sleep, knight festival, lunch, dinner, and play) built with

Mario Zechner

67,621 views • 11 days ago

sneak peak of the new no tools mode. you're gonna love it. you can try it with `pi -nbt`. enjoy!
1:04
badlogicgames's profile picture

sneak peak of the new no tools mode. you're gonna love it. you can try it with `pi -nbt`. enjoy!

Mario Zechner

112,345 views • 20 days ago

now doing gemma 4 moe. it'a crazy how usable this is. just a little web search tool (wiggles its head on tool calls), and it can serve as a mostly capable, local voice assistant. tomorrow i shall replace google auto crap with this in my car. just need to give it access to spotify.
1:44
badlogicgames's profile picture

now doing gemma 4 moe. it'a crazy how usable this is. just a little web search tool (wiggles its head on tool calls), and it can serve as a mostly capable, local voice assistant. tomorrow i shall replace google auto crap with this in my car. just need to give it access to spotify.

Mario Zechner

27,123 views • 6 days ago

how it happened. Armin Ronacher ⇌: "sensai, the code, it sparks no joy anymore." pidalf: "don't be a such a god damn fool." Armin Ronacher ⇌: "thank you. i can see the light again."
0:36
badlogicgames's profile picture

how it happened. Armin Ronacher ⇌: "sensai, the code, it sparks no joy anymore." pidalf: "don't be a such a god damn fool." Armin Ronacher ⇌: "thank you. i can see the light again."

Mario Zechner

39,638 views • 14 days ago

People of pi has supported OpenAI WebSockets mode since February. However, we did not support delta updates. pi v0.71.1 now also supports delta updates: only the latest context additions get send, leading to a nice 66% throughput increase. Start pi, /settings -> transport -> websocket-cached Here's SSE vs. cached WebSocket side by side. This will also apply to your OpenClaw instance, should Peter Steinberger 🦞 & team decide to update to the latest and greatest, giving you all the benefits of the pi runtime on top of the Codex server backend.
0:43
badlogicgames's profile picture

People of pi has supported OpenAI WebSockets mode since February. However, we did not support delta updates. pi v0.71.1 now also supports delta updates: only the latest context additions get send, leading to a nice 66% throughput increase. Start pi, /settings -> transport -> websocket-cached Here's SSE vs. cached WebSocket side by side. This will also apply to your OpenClaw instance, should Peter Steinberger 🦞 & team decide to update to the latest and greatest, giving you all the benefits of the pi runtime on top of the Codex server backend.

Mario Zechner

68,881 views • 1 month ago

4yo used my little TTS desktop app + pi, tells the agent what to build, what to change. more dinosaurs to select from, now working on king kong as final boss. i love this so much. slop doesn't matter. early agency does. and he gets to understand on the highest possible level how things are made.
0:32
badlogicgames's profile picture

4yo used my little TTS desktop app + pi, tells the agent what to build, what to change. more dinosaurs to select from, now working on king kong as final boss. i love this so much. slop doesn't matter. early agency does. and he gets to understand on the highest possible level how things are made.

Mario Zechner

55,588 views • 1 month ago

Here we see GPT-5.1 struggle to read a 500 lines markdown file for almost 2 minutes. Same model in a different harness does it in a few seconds (takes 2 tool calls for whatever reason..). Is all the Codex-love an elaborate joke I'm not in on?
2:23
badlogicgames's profile picture

Here we see GPT-5.1 struggle to read a 500 lines markdown file for almost 2 minutes. Same model in a different harness does it in a few seconds (takes 2 tool calls for whatever reason..). Is all the Codex-love an elaborate joke I'm not in on?

Mario Zechner

209,999 views • 6 months ago

The most hilarious part is that Claude Sonnet 4.5 is a gazillion times faster this way compared to using it via CC... The stats displayed in pi on the left are accumulative across all requests, while the context stats in CC are only for the last message.
1:21
badlogicgames's profile picture

The most hilarious part is that Claude Sonnet 4.5 is a gazillion times faster this way compared to using it via CC... The stats displayed in pi on the left are accumulative across all requests, while the context stats in CC are only for the last message.

Mario Zechner

194,911 views • 6 months ago

Sloppy tools for sloppy times: doppelgangers Explore all the slop issues and PRs of your GitHub repos visually. Triage them manually or give clusters to your clanker to clean things up. Here's OpenClaw🦞
1:04
badlogicgames's profile picture

Sloppy tools for sloppy times: doppelgangers Explore all the slop issues and PRs of your GitHub repos visually. Triage them manually or give clusters to your clanker to clean things up. Here's OpenClaw🦞

Mario Zechner

112,099 views • 4 months ago

Putting my tokens where my mouth is. I built pi-share-hf. Share your pi coding agent sessions as Hugging Face datasets. It tries to prevent you from uploading sessions containing PII/sensitive data with 3 tiers of defenses. Best used on OSS coding sessions, as those are less likely to contain sensitive info. Uses pi agents for PII detection, which can cost you a lot of tokens. Read the README with your human eyes so you don't accidentally pwn yourself or get a huge bill. Haven't figured out how to filter for such datasets on HF yet. clem 🤗, any pointers on how to best label them so people can find them?
12:09
badlogicgames's profile picture

Putting my tokens where my mouth is. I built pi-share-hf. Share your pi coding agent sessions as Hugging Face datasets. It tries to prevent you from uploading sessions containing PII/sensitive data with 3 tiers of defenses. Best used on OSS coding sessions, as those are less likely to contain sensitive info. Uses pi agents for PII detection, which can cost you a lot of tokens. Read the README with your human eyes so you don't accidentally pwn yourself or get a huge bill. Haven't figured out how to filter for such datasets on HF yet. clem 🤗, any pointers on how to best label them so people can find them?

Mario Zechner

51,391 views • 2 months ago

People who like sharing agent traces. I've just published all my pi-mono coding agent sessions on Hugging Face so you get to laugh at or pwn me! I suggest you do the same, see thread below. Let's make this a community effort. Here's pi-share-hf: If you are working on tools that help identify PII/sensitive data, get in touch. The better the classification is, the more willing people will be to share their traces.
7:29
badlogicgames's profile picture

People who like sharing agent traces. I've just published all my pi-mono coding agent sessions on Hugging Face so you get to laugh at or pwn me! I suggest you do the same, see thread below. Let's make this a community effort. Here's pi-share-hf: If you are working on tools that help identify PII/sensitive data, get in touch. The better the classification is, the more willing people will be to share their traces.

Mario Zechner

45,750 views • 1 month ago

A new entry to my popular series "LLM tools for plebs": claude-trace - Injects itself into Claude Code - Logs all traffic - Reconstructs conversations and shows what's going on behind the scenes (system prompts, all tool inputs/outputs, and more) Some observations. 🧵
2:10
badlogicgames's profile picture

A new entry to my popular series "LLM tools for plebs": claude-trace - Injects itself into Claude Code - Logs all traffic - Reconstructs conversations and shows what's going on behind the scenes (system prompts, all tool inputs/outputs, and more) Some observations. 🧵

Mario Zechner

221,359 views • 1 year ago

Every now and then I try GitHub Copilot in VS Code. I truely wonder if anyone working on it is actually using it. The UX is mindboggling bad. Hide session bar, chat area gets small, resize chat area, session bar appears again. Make it make sense.
0:20
badlogicgames's profile picture

Every now and then I try GitHub Copilot in VS Code. I truely wonder if anyone working on it is actually using it. The UX is mindboggling bad. Hide session bar, chat area gets small, resize chat area, session bar appears again. Make it make sense.

Mario Zechner

83,425 views • 5 months ago

if an agent emits parallel tool calls, pi used to execute them sequentially. funnily enough, only a handful of people complained. welp, just implemented it. caveat: you'll get flicker if you have many parallel tool calls and their outputs are > terminal height. soon to be released.
0:51
badlogicgames's profile picture

if an agent emits parallel tool calls, pi used to execute them sequentially. funnily enough, only a handful of people complained. welp, just implemented it. caveat: you'll get flicker if you have many parallel tool calls and their outputs are > terminal height. soon to be released.

Mario Zechner

35,235 views • 2 months ago

People of pi. Cloudflare release with a minor breaking change. - JSON validation no longer relies on AJV, but uses Typebox. Update your existing extensions as per changelog (@sinclair/typebox -> @typebox) - OSC 9;4 indicator in supporting terminals for Lucas Meijer thanks to MVP Kayla Cinnamon ☕ - Tools can now signal "terminate the loop" so you can do structured output and other shenanigans easily - Custom auto completion providers via extension. This one fetches the last 100 isues from GH when i type # and let's me search through them via type ahead And lots of bug fixes and internal refactoring.
0:42
badlogicgames's profile picture

People of pi. Cloudflare release with a minor breaking change. - JSON validation no longer relies on AJV, but uses Typebox. Update your existing extensions as per changelog (@sinclair/typebox -> @typebox) - OSC 9;4 indicator in supporting terminals for Lucas Meijer thanks to MVP Kayla Cinnamon ☕ - Tools can now signal "terminate the loop" so you can do structured output and other shenanigans easily - Custom auto completion providers via extension. This one fetches the last 100 isues from GH when i type # and let's me search through them via type ahead And lots of bug fixes and internal refactoring.

Mario Zechner

20,650 views • 1 month ago

I present to you my latest fully vibeslopped "I didn't read the code" oeuvre: Jot A self-hosted collaborative markdown editor for you, your friends, and your agents. Like Google Docs, but worse. Try it with me here: Share a note, pick access (view/comment/edit), hand the link to a human or paste the CLI instructions into your agent. They can read, edit, and comment together in real-time.
5:31
badlogicgames's profile picture

I present to you my latest fully vibeslopped "I didn't read the code" oeuvre: Jot A self-hosted collaborative markdown editor for you, your friends, and your agents. Like Google Docs, but worse. Try it with me here: Share a note, pick access (view/comment/edit), hand the link to a human or paste the CLI instructions into your agent. They can read, edit, and comment together in real-time.

Mario Zechner

28,409 views • 2 months ago

Remember my funky browser agent? Well, decided to make it free and OSS. It did summarize YouTube videos before it was cool! OSS will take a bit of clean-up. If you can't wait, you can already install it from here:
0:41
badlogicgames's profile picture

Remember my funky browser agent? Well, decided to make it free and OSS. It did summarize YouTube videos before it was cool! OSS will take a bit of clean-up. If you can't wait, you can already install it from here:

Mario Zechner

49,584 views • 5 months ago

This is a very nice description of the problem. I think the first part of the solution is to not give into the hyper productivity hype: running a gazillion things in parallel. There's likely only a handful of humans that have the mental make-up to survive the permanent context switches. You can train your brain to become better at it, but there's a limit to that, and I found that it will lead to much worse outcomes in terms of output quality of both you and your clanker. I'm no neuroscientist, but I would assume there's a hard limit on how many things your brain can juggle at a time, probably made worse by the complexity that is code. I generally don't do more than 2-3 things in parallel. I use virtual desktops as "bundle of things that belong to a task", so terminals + editor + browser. Works fine, doesn't need any special OS support, expose on the virtual desktop let's me quickly switch to the task specific app I need. I also have a little pi extension that shows me what GH issue/PR the clanker session is about, with a link I can click to open the issue/PR in the browser. Plus 1 virtual desktop for entertainment, so the brain can get some rest. (better alternative: get up and walk a bit every 45 minutes) Video: - First desktop: pi session for a PR, extension shows me what PR I work on, can click to open in browser, aux terminal for random BS, editor, browser. Yes, I could split instead of having two terminal windows. I'm old. - Second desktop: clawdbot pi session, aux terminal for testing, editor, browser - Third desktop: entertainment booth
0:23
badlogicgames's profile picture

This is a very nice description of the problem. I think the first part of the solution is to not give into the hyper productivity hype: running a gazillion things in parallel. There's likely only a handful of humans that have the mental make-up to survive the permanent context switches. You can train your brain to become better at it, but there's a limit to that, and I found that it will lead to much worse outcomes in terms of output quality of both you and your clanker. I'm no neuroscientist, but I would assume there's a hard limit on how many things your brain can juggle at a time, probably made worse by the complexity that is code. I generally don't do more than 2-3 things in parallel. I use virtual desktops as "bundle of things that belong to a task", so terminals + editor + browser. Works fine, doesn't need any special OS support, expose on the virtual desktop let's me quickly switch to the task specific app I need. I also have a little pi extension that shows me what GH issue/PR the clanker session is about, with a link I can click to open the issue/PR in the browser. Plus 1 virtual desktop for entertainment, so the brain can get some rest. (better alternative: get up and walk a bit every 45 minutes) Video: - First desktop: pi session for a PR, extension shows me what PR I work on, can click to open in browser, aux terminal for random BS, editor, browser. Yes, I could split instead of having two terminal windows. I'm old. - Second desktop: clawdbot pi session, aux terminal for testing, editor, browser - Third desktop: entertainment booth

Mario Zechner

40,184 views • 4 months ago

People of pi. I'm happy to announce pi 0.47.0 which brings: npm install -g @mariozechner/pi-coding-agent - Official OpenAI Codex support. Use your OpenAI Plus or Pro plan (thanks Tibo) - IME support for our CJK friends - Many fixes, improvements, and new features
1:13
badlogicgames's profile picture

People of pi. I'm happy to announce pi 0.47.0 which brings: npm install -g @mariozechner/pi-coding-agent - Official OpenAI Codex support. Use your OpenAI Plus or Pro plan (thanks Tibo) - IME support for our CJK friends - Many fixes, improvements, and new features

Mario Zechner

39,673 views • 4 months ago

People of pi. Welcome to the meta coding agent harness 0.50.0. npm install -g @mariozechner/pi-coding-agent - Bundle your extensions, skills, prompt templates, and skills as pi packages, distribute via NPM or git - Hot-reload everything via /reload A 🧵 of possibilities (1/x)
0:25
badlogicgames's profile picture

People of pi. Welcome to the meta coding agent harness 0.50.0. npm install -g @mariozechner/pi-coding-agent - Bundle your extensions, skills, prompt templates, and skills as pi packages, distribute via NPM or git - Hot-reload everything via /reload A 🧵 of possibilities (1/x)

Mario Zechner

36,043 views • 4 months ago