Kamen Dimitrov

CTO @ Ordinary Objects (

[Kamen Dimitrov @kamend]

veeery relevant rigjt now

[vicki @vboykis]

new post: how I develop recently using local models. the tooling is now good enough to do agentic workflows and everyone should give them a try!

https://t.co/3Tx3CMsNG3

[Kamen Dimitrov @kamend]

@pvncher Absolutely, i was very sceptical about this one in particular

[Kamen Dimitrov @kamend]

@nuvolore This is a very clever approach, thanks for sharing!

[Kamen Dimitrov @kamend]

@omarsar0 wow, this is great, thanks for sharing.. there goes my sunday

[Kamen Dimitrov @kamend]

@mattshumer_ @mrdoob where all the 3D models came from?

[Kamen Dimitrov @kamend]

https://t.co/yqLxMhPd8K

[Kamen Dimitrov @kamend]

Research shows that 26% of Agent Skills contains vulerabilities and 5% show likely malicious intent, NVIDIA just released a security scanner that helps you check if a skill is safe to install!

[Kamen Dimitrov @kamend]

https://t.co/tm6rtY1tGX

[Kamen Dimitrov @kamend]

You might be thinking that it has been a while and LLM models are more secure and safe from prompt injections and data exfiltration attacks @wunderwuzzi23 has bad news for you all. Read the article in the reply!

[Viv @Vtrivedy10]

imo there’s a pretty solid default recipe that everyone should use to optimize a system of
Agent = Model + Harness

you should “train” both

1. Build v1 agent using a sensible base harness and some task specific prompting + tools

2. Harness Engineering using eval tasks that roughly match prod

this is often enough - most companies can get acceptable perf doing this. then they collect traces, mine them for patterns, and make slight tweaks from there

3. SFT using data collected from traces) or synthetic data. Often is good candidate for “distillation tasks” to train a cheaper model while maintaining existing performance

4. RL if you have the bandwidth and ability and desire to create environments and designing rewards that represents the tasks you want your agent to be good at. Push past the SFT behavior of “copying” data from existing model to pushing past in some dimension

5. Light harness engineering again to squeeze any more juice (ex: slight prompting) using the trained model that’s better at your task distribution

this loop will largely be productized as a general purpose recipe for building and improving agents

we’re still in the earliest innings of the world’s companies getting comfortable with steps 1-2 of this loop. Harness engineering will probably be the dominant way ppl will optimize agents

but i expect a large number of companies to onboard through this entire loop on some trial project of interest in the next year

[0xkato @0xkato]

LLMs explained without all that yucky math stuff

https://t.co/cpgVukSbJx

[Kamen Dimitrov @kamend]

@divaagurlxw tip: give this to Claude Code and tell it generate Python exercises and samples for all of it, if you like to learn by practice like me

[Kamen Dimitrov @kamend]

@galluzzo_julian let’s see.. they are going to release.. an agent 😀

[Kamen Dimitrov @kamend]

https://t.co/hT75DuIDi3

[Kamen Dimitrov @kamend]

No idea why I am just finding out about llmfit, great tool to find all the LLM models you can run locally on your machine! #llm #ai

[⚡AI Search⚡ @aisearchio]

NVIDIA's LocateAnything is a new vision model for grounding and detection. Very performant and accurate!

> 10x faster than Qwen3-VL
> 138M queries + 785M boxes
> GUI, OCR, docs, dense detection
> Free & open source

https://t.co/UvkH8l0QRb

[ClaudeDevs @ClaudeDevs]

We’ve shipped a security-guidance plugin for Claude Code that helps identify and fix vulnerabilities as you’re writing code.

Available for all Claude Code users. Install from the plugin marketplace (/plugins).

[Kamen Dimitrov @kamend]

@and1truong down for me too, what a surprise this morning

[Mario Zechner @badlogicgames]

recommended reading. also read part II. also read naur's original article.

https://t.co/O6pe3lTxix

[Kamen Dimitrov @kamend]

https://t.co/qSrirHXTB6

[Kamen Dimitrov @kamend]

A 16-line bash script that builds its own coding agent harness, then uses the harness to build a backdoor file browser. No frameworks. No dependencies beyond `curl`, `jq`, and `python3`

A simple prompt and a network call and you have a T-1000 kind of “weapon” at your disposal.