@RebentischT @bridgemindai Actually, Now that I look at the numberd again on openrouter, the prices are identical across providers... I don't know what I was looking at...
English
Fran
2K posts
Kimi K3 open weights are live and it is already running 3x faster.
The weights dropped this morning. Fireworks is already serving it at 45 tokens per second with 1.44s latency.
I have spent 11 days waiting 11 seconds for this model to say its first word. It was painful. The best open model ever made and I kept closing the tab.
Not anymore.
This is the part of open source people forget. You release the weights and the entire industry fixes your problems for you. Same day.
Kimi K3 just became a real vibe coding option.
English
Ordered a chorizo salad the other day
Got a chorizo sandwich
I thought "this feels familiar"
English
@DanielGlejzner All harnesses support working across directories / projects / workspaces, call it what you want.
So if that's the main reason, it's no longer relevant. Used to be true months ago.
English
Monorepo is going to be the most optimal setup to properly work with AI given all the context is in one codebase
English
Starting with Gemini 3.6 Flash and 3.5 Flash-Lite, temperature, top_p, and top_k are deprecated.
learn more in our dev guide: #api-changes-and-parameter-updates" target="_blank" rel="nofollow noopener">ai.google.dev/gemini-api/doc…
English
@ericlancheres This tool needs to SSH into web server?
It doesn't just access the website and crawl it by it's public URL?
How would I for example audit my next.js website that's hosted on Vercel or Netlify?
English
It can connect to any website that has SSH enabled (so pretty much every website works).
The cool thing about Wordpress is that there's a wizard where you just put in the site and it uses Chrome to set things up for you so it makes things really easy... which is why I was showing it off!
English
Introducing the GEO SEO SuperApp, an open-source desktop app for SEO & GEO agencies.
I built it because I was tired of using multiple apps and wanted everything in one place.
Pick a recipe, enter the URL and watch it work.👇
English
Can someone please explain what is thaaat special about this, that people say it's a historic moment?
It's pretty cool but basically you control Codex in a call?
(I may be completely wrong and it could be something CRAZY, but I don't get it yet)
OpenAI@OpenAI
ChatGPT Voice is now in the desktop app. Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice. It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time. Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.
English
@JackWoth98 How does this work?
I see he runs agent.py script and I assume he uses the interactions API to call the agent.
This means you need to configure 'computer use' tool somehow manually. No?
How is that done?
Can it control my laptop?
English
Gemini 3.5 Flash-Lite now has Computer Use! 🖱️
This is the first time a Flash-Lite model has the tool.
Let the speedy model control your browser, computer, or phone 🖥️📱
I think there are a lot of cool unlocks with this combo!
Patrick Loeber@patloeber
fyi, gemini-3.6 flash & 3.5-flash-lite support the Computer Use tool across browser, desktop, and mobile. here’s a demo of Flash-Lite operating Android. we also shipped a GitHub repo so you can get started easily:
English
Grok Build gets better every day
X.ai/cli
X Freeze@XFreeze
Grok Build just got major update, giving developers more control over AI tools, better account visibility, built-in diagnostics, stronger session reliability, and lower memory usage for Voice Mode on macOS. Release Notes: v0.2.111 Features: • Users can now disable image generation and video generation tools (and their slash commands) via config.toml or environment variables. • /session-info now displays whether the session uses OAuth or an API key and where to manage the account. • You can now run grok doctor fix commands directly from inside the TUI instead of only from the CLI. Bug Fixes: • !cmd commands now allow up to one hour before timing out. • npm package now installs the native binary under $GROK_HOME/bin (honoring the same override as the Rust CLI). • Startup warnings now point to /doctor for details and fixes. • Dashboard hover and clicks no longer miss the gaps between items in wide mode. • Shift/Alt+Enter now inserts a newline while editing a queued prompt. • Queued prompt edits under combine mode no longer lose changes due to premature hold release. • Forking a session that used compaction no longer causes later rewinds to fail with missing checkpoint errors. • When a permission prompt appears while viewing scrollback, focus now correctly moves to the prompt so you can answer. • Pressing Esc once now cancels the current agent turn (except in fullscreen vim scrollback mode). • Grok now automatically stops a turn that keeps repeating the exact same tool call many times in a row. • Configs using either spelling of the workspace teleport disable flag now load and save correctly. • Background subagent completion messages no longer leak into unrelated sessions when multiple sessions are active. • When the auto-permission classifier times out or fails, Grok now shows a normal permission prompt instead of silently denying. • Managed MCP tools no longer time out prematurely on slow operations like Notion updates. Performance: • Voice dictation on macOS now uses less memory by running capture in a temporary helper process.
English
@Taniyatweets_ @AlfinCodes This, people don't care, and they use what they got their hands on first.
English
@mattpocockuk tldr: build a wayfinder canvas that Ai agents can interact with like they would with gh cli to retrieve/query state of the wayfinder session or tickets, while still having a custom UI for the user as well.
English
@mattpocockuk I'm not sure what the purpose of a custom harness for wayfinder would be, but if it has to do with management and presentation of tickets, I would recommend you look into builderio/agent-native and their /visual-plan skill.
English
It's on my list to ship my own Pi-powered harness based on Wayfinder
I just have like 6 things to ship first
English
@opencode @metruzanca Two more whys:
1. Why can’t I autocomplete skills mid-prompt?
2. Why doesn’t the submit button have a hover style?
English
@metruzanca You can turn this on in the settings but with the newer models you don't typically need a separate plan mode
English
@FranVidiCheck github.com/agentSystemLab… it's open source ask an agent over the code
English
At this point I've dropped all other ADE tools and made my own. None of them are designed for working on multiple projects efficiently. I added the ability to archive sessions months ago.
Theo - t3.gg@theo
I give it 10 days before everyone starts copying this workflow. Seriously can't believe we've been dealing with single rows of text for over a year.
English
@MatthewBerman What is the rationale behind using a fast (implied dumber) model for execution?
Plans are usually not so detailed that will prevent bad decisions that fast model will make when writing code.
If they were that detailed, why not implement immediately instead of writing the plan..
English
Big Relay update!
Relay is an open-source skill for routing code writing through multiple models: frontier planning → fast execution → frontier review.
Save money on tokens, increase speed and quality.
Now works in Codex, Cursor + Claude Code, discovers models, knows frontier from workhorse, and catches silent fallbacks.
github.com/Forward-Future…
English
@MatthewBerman @PhilP2874 Number of tasks sucessfully completed is quite important.
English
The main metric that matters is price per task completion.
Second is speed to task completion.
Nothing else matters.
English
