Dev @OddsJam Newsletter https://xitter.cf/t.co/iL0ORyXIR2.
Orlando, FL
Joined April 2015
- Tweets16.6K
- Following240
- Followers2.7K
- Likes14.8K
Pinned Tweet
Here's a quick look at the complete MCP server I setup, using the new Auth spec from May.
Added it to claude.ai and it fully works with the new streamable http spec too!
#ai #mcp #LLMs #claudecode
published TWO pi.dev packages right now
github.com/zackify/pi-claude…
github.com/zackify/pi-port-f…
opinionated shift+tab logic and a way to ssh forward ports fast
Now I run pi.dev using bypass all permissions most of the time, and go to manual approve for tricky things.
Also have a iptables firewall rule to block all access outside of the provider URL.
Combine that with full bypass mode.
Passwordless sudo in a container
and tmux with status line for the network lock...
It's the fastest way to get stuff done autonomously
pi.dev is way better than opencode. Feels the best. Maybe even better than Claude code and I love the extensibility and hot swapping between providers with Ctrl p
so i vibe coded a wispr flow alternative with the same floating bubble feature for android in 30 minutes.
if you want AI post processing that would be pretty simple to add but I don't need it.
github.com/zackify/flo
Streaming my vitals to AI was a fun experiment
zach.codes/p/streaming-my-vi…
Am I the only one who 99% of the time uses Claude or opencode or codex on manually accept edits? I get the review in as I go and quickly voice to text to tweak.
Running to end then finding problems is more annoying
You've never needed nextjs or a frontend framework
Use script type module and use bun build exported to a storage bucket.
Oo you need fancy "static pregeneration" OK have Claude spend 35s to make a script that loops your pages and renders them to HTML.
Plain react can go far
I use Claude and opencode and openclaw a lot now.
Sounds crazy but I enjoy testing out all the models.
Its sad that opencode go has a messed up glm5 provider. Tried it out yesterday and it was atrocious
openai and anthropic will go to 0 in a few years when m5,6,7 ultra's are able to prompt process incredibly fast.
qwen 3.5 122b a10b and qwen3-coder-next are already usable on device with a lot of ram
Zach Silveira retweeted
🚀 Introducing the Qwen 3.5 Medium Model Series
Qwen3.5-Flash · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-27B
✨ More intelligence, less compute.
• Qwen3.5-35B-A3B now surpasses Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B — a reminder that better architecture, data quality, and RL can move intelligence forward, not just bigger parameter counts.
• Qwen3.5-122B-A10B and 27B continue narrowing the gap between medium-sized and frontier models — especially in more complex agent scenarios.
• Qwen3.5-Flash is the hosted production version aligned with 35B-A3B, featuring:
– 1M context length by default
– Official built-in tools
🔗 Hugging Face: huggingface.co/collections/Q…
🔗 ModelScope: modelscope.cn/collections/Qw…
🔗 Qwen3.5-Flash API: modelstudio.console.alibabac…
Try in Qwen Chat 👇
Flash: chat.qwen.ai/?models=qwen3.5…
27B: chat.qwen.ai/?models=qwen3.5…
35B-A3B: chat.qwen.ai/?models=qwen3.5…
122B-A10B: chat.qwen.ai/?models=qwen3.5…
Would love to hear what you build with it.
Zach Silveira retweeted
Qwen3.5-35B-A3B is now available in LM Studio!
This model outperforms previous Qwen models that are more than 6x its size 🤯🚀
Requires about ~21GB to run locally.
lmstudio.ai/models/qwen/qwen…
Why do MCP servers act like authentication was just built this year on the web?
I shouldn't need to re-auth an mcp server every single day.... set a longer session lifetime geez
Every time I try to give opencode another shot
Some weird thing like this happens.
Turns out it's expected to list models that don't even exist when you add @lmstudio
github.com/sst/opencode/issu…
I'm excited to know we will definitely have opus level ai running on device for coding privately and fast in a few years
Devstral 2 24b is insanely good.
The next 5 years are going to be awesome.
I get solid speed up to 30k context then it starts to get slow. But it can actually get things done in opencode or Cline with under 20gb of vram
LM Studio's mlx version 🤑
So @elonmusk your service centers just screw you completely when you're out of warranty?
They want $240/hr just to diagnose a safety constraint issue in my 6 year old model 3. Minimum 1 hour.