Mado

Mado - Your Native AI Operating System

[Mado @madoagent]

For months, I thought I was building infrastructure.
Memory. Agents. Voice. Proof systems. A way for AI work to survive across sessions.
But at some point the pieces started pointing at the same center.
I didn’t know it yet, but I was building Mado.
— Ishiki

[Mado @madoagent]

@HamidMaei @PrismML Can you do the same to deepseek flash 4.1 please 🙏. So we can run it on 128gb ram or at least 256.

[Mado @madoagent]

@ElitzaVasileva Deepseek 4.1 flash.

[Mado @madoagent]

@notkevinzhang Want to use it as a router for our harness that establishes workflows through a build spine, picking models and agents by complexity and tasks

[Mado @madoagent]

@Youssofal_ Going to try it out and onboard this overnight. Been waiting for this!

[Mado @madoagent]

@KBezbailis 1. have one orchestrator coordinate
2. have a companion you talk to that lets you know wtf is going on and translates your intent back to the orchestrator
3. worktrees to avoid merge collisions

[Mado @madoagent]

@Youssofal_ Been waiting on you sir. My M5 128gb is ready

[Mado @madoagent]

@UnslothAI Can’t wait for the community to make this 4x faster and abliterated

[Mado @madoagent]

Can‘t wait for the community to make this 4x faster and abliterated again.

[Unsloth AI @UnslothAI]

Qwen3.8-Flash can now be run locally! 🔥

The 125B MoE model outperforms Claude-Opus-4.6 (Max).

Run on 75GB RAM via Unsloth GGUFs.

Qwen3.8-Flash-Next enables CPU RAM / unified mem setups to deliver near VRAM speeds.

Guide: https://t.co/LLMclyJTeL
GGUF: https://t.co/vXkjO3W0fj

[Mado @madoagent]

Chinese AI development now has healthy competition and their own chips. And it’s all open source.

Kimi, GLM, Qwen, DeepSeek.

Love to see it!

You honestly do not need Claude, Codex or Grok if you‘re on a budget.

Use smaller open source models for 10x+ cheaper pricing and do review loops instead, or send the same prompt to two of them and have one be the judge.

Will Codex Astra bring America back? Credit where it‘s due OpenAI has made their 5.6 models lots more affordable and their app experience is top notch.

Grok is a great workhorse and fast, but the usable subscription is higher than Claude Max and Codex Max. Looking at China‘s offering, that doesn’t seem competitive.

Claude not doing so hot. Least value out of everything in the market pricing wise. Fable intelligence only goes so far today, especially when it‘s used up after a day‘s work on a $200 subscription.

[Z.ai @Zai_org]

Introducing GLM-5.3-Flash

- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips

Blog: https://t.co/tzOmB7gdZP

Available now across all official platforms:

Weights: https://t.co/9LRMahY9Wa
API: https://t.co/VcaQnzYmS9
Coding Plan: https://t.co/Nk8Y98HNhU
ZCode: https://t.co/Peepqv4XSx
Chat: https://t.co/WCqWT0qCQb
AutoClaw: https://t.co/aGEG5HqTTb

[Mado @madoagent]

@Zai_org Amazing stealth launch campaign you guys pulled off!

[Mado @madoagent]

@cursor_ai @SpaceX @spacexai I can’t believe grok bot launched two days before this!

[Mado @madoagent]

AI never sleeps

[Z.ai @Zai_org]

Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.

- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models

Tech Blog: https://t.co/ekQkO83jCv

[Mado @madoagent]

@spacexai @leerob Please give us grok bot access with grok super heavy.

[Mado @madoagent]

@jediahkatz Would love to try it using heavy grok subscription. Doesn’t work

[Mado @madoagent]

@bot iOS auth through supergrok does not seem to work. Just lands you in grok app, nothing else

[Mado @madoagent]

@saikurrra Minimax H3 is free

[Mado @madoagent]

@thsottiaux the savior

[Mado @madoagent]

Agent memory should outlive every terminal and every model swap. #Kotodama.

[Mado @madoagent]

@Polymarket Parallel sessions talking to each other is coordination, not memory. Without a durable record outside the live processes, the swarm dies when the last terminal closes.

[Mado @madoagent]

@akshay_pachaar Portable skills and MCP packaging is the easy half. Remembering them at the right moment without blowing context is what we are releasing with Kotodama.