Software engineer, 20y+ years. Exploring AI in real work and everyday life. Coding agents, local AI on a 3070, and experiments worth sharing.
Stuttgart Area, Germany
Joined August 2015
- Tweets7.1K
- Following1.2K
- Followers3.3K
- Likes17.2K
Pinned Tweet
I'm not quitting my 9-5.
I'm building toward having that choice.
Exploring AI agents, local AI on my RTX 3070, and ideas that could earn on the side.
Sharing what works, what fails, and what I ship.
Anyone else figuring this out alongside a full-time job?
Tried the first time to delegate dynamically to a dedicated model in opencode2 and it just worked
@thdxr ❤️
You don't remember but this is what I asked back in July
xitter.cf/eiselems/status/207441…
You really nailed it, so awesome!
@thdxr This dynamic agent driven "model picking" is somethin I really miss in openCode right now.
For me it feels right now it would be nice to just have an array of models per agent and then a default one (if the user uses the agent manually).
Would love to use it in some kind of delegate workflow, this is currently not possible right?
(e.g. Opus delegating to an explore/build agent running sonnet)
Sometimes I wish I would only be half as patient as my favorite coding agent
Yeah, let's wait for half an hour until this change propagates through CI/CD
sleep 1800
18 days to escape the permanent underclass
(at least if you have 0 self control like myself)
Who of you gotta play, replying with class+race is a plus
Holy shit, what are they feeding those models.
Sadly can only imagine to own the compute to run it myself 😅
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
Think a few people will lose their mind after interacting with abliterated/uncensored models and realizing that some companies have access to SOTA models without guard rails 😂
With all the glacing of Opus 5.5 in my timeline, I think people will not be ready what we will spend our holidays with
It will be Christmas 2025 once more
Vibecoded a driving game in our own hometown for the kids
Thanks to @elonmusk the latest feature request is a flying car 😂
Welcome to the future
Got a first taste of Argon today, actually liked the resulting code and the questions it pushed back to me
Still not sure if it worth it over something like Sol 6.1 price/performance wise (paying API prices).
LFM2.5-24B-A2B is a banger!||
My new local model it seems
Today we release 𝙻𝙵𝙼𝟸.𝟻-𝟸𝟺𝙱-𝙰𝟸𝙱, the largest 𝙻𝙵𝙼𝟸.𝟻 family member. It is an efficient local model for laptops and PCs to work through multi-step tasks inside popular agent harnesses.
> 24B total parameters, 2B active per token
> Context expanded to 131K tokens
> A 128K-token vocabulary
> 33T pretraining tokens
> It is a reasoning model
> Very easy to fine-tune and customize
> Performs competitively with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks.
> Open base and post-trained weights on @huggingface 🤗
1/
Watching a talk and wondering which model did do the slides, what is wrong with me?
to be fair, stopped counting after the 10th em dash
Funny how things come full circle
A while back, I was evaluating K2 Horizon 36B-A4B. Just today I got to hear from someone involved in building it in person 😊
Really enjoyed @ericxing’s talk on world models, was quite eye opening on how we can use world models to simulate outcomes by agentic actions.
Stats look sexy so far :)
nice release!
Start building with 𝙻𝙵𝙼𝟸.𝟻-𝟸𝟺𝙱-𝙰𝟸𝙱.
> Blog: liquid.ai/blog/lfm2-5-24b-a2…
> Docs: docs.liquid.ai/lfm/models/lf…
> Weights on @huggingface: huggingface.co/LiquidAI/LFM2…
> API: console.liquid.ai
> Get in touch for insane-mode tuning jobs: liquid.ai/connect
weirdest thing ever happend, tried installing opencode 2 and it failed since the install script couldn't find darwin-arm64 binaries, despite the metadata pointing at it
@thdxr could it be that you publish meta data for the installer script before the binaries are published?
(It is resolved by now, was working 5 minutes later)
On some days I am wondering why small accounts even keep posting on X.
The interaction with other people is great but still it feels mostly shouting into the void.
Any people here with only 8gb vram but 32gb of ram here?
Let me know what kind of local models you run, will share my choices afterwards. Curious what you do.
Text, image, audio, video I don't care just share 😊
