Alex Dunne

@thekitze this. I start with opencode and then build from there. nice web ui out of the box and solid DX. much more polished and simple

what are your usecases for openclaw/hermes that you cannot do with codex / claude code (local or remote via phone) I am legit asking and not trolling.

@ShubhamMal72313 @RunAnywhereAI @liquidai make openclaw-type personal assistant app using this?

Early beta: @RunAnywhereAI Web SDK Build LLM / Voice / Vision apps + tool calling - straight in the browser. Also: this demo uses a @liquidai VLM (small + fast). Huge shoutout to the Liquid team 🙌 Try it: runanywhere-web-demo.vercel.app #runanywhere #yc #websdk #vlm #llm #voice #localai

who's working on openclaw mobile? seems like a natural fit since modern smartphones can run sufficiently capable models locally, are always on, and are connected to everything. i bet that's what we'll see within 3-6 months from openai

intimate dinners are a weirdly effective funnel. In niche older industries, a lot of deals and long term relationships still only move forward face to face. we’ve been testing this recently and the signal has been interesting to observe. people leave with a much clearer understanding of your motive

@jai_mansukhanii this tweet perfectly encapsulates the autistic side of our tech bro culture hahaha

@BraydenWilmoth imo a lot of that is dependent on good STT to steer via phone without having to type. been playing around with parakeet TDT v3 and its been great. does cf have support for that?

My Q1 goal is to develop an AI agent that does my job for me – while I'm on the beach – and I'm close! I add my agent to calendar events, it joins standups (Browser Rendering + Realtime Kit), listens to participants (moves Jira tickets + take notes), enriches ticket descriptions, moves tickets based on PR statuses, listens to Sentry errors and logs & prioritizes them... It does a lot of the boring parts of running a team so I can focus on the more interesting problems to be solved (like building more agents).

clawdbot + opencode web = phone-driven development (PDD) @openclaw

@arun_sankr @steipete @eliasstravik @openclaw did the exact same yesterday too lol github.com/adunne09/suit

@steipete @eliasstravik @openclaw Added one yesterday clawdhub.com/skills/linkedi…

won't be trying @openclaw until its available on linkedin smh rip @steipete

@0xblacklight yes

Is clawdbot that good

codex is a beast, holy shit. give it the right tools for the job and a nudge to self verify with those tools and it just goes. makes me curious how a ralph-style loop with a weaker|faster|cheaper model compares

@kristianfreeman why tailscale vs cloudflared/cf tunnel?

The migration from my VPS to my local computer wasn't great! Basically had to start from scratch. But now it's working wonderfully - web UI is available via Tailscale, and everything is locked down on my local network. Worth it!

deep work reports and streaks-

was inspired by @jackfriks deepworkdepot.com showing reports in gh-style contribution graphs so i made one for @openclaw in the last 20 minutes clawdhub.com/adunne09/deepw…

@steipete what's next for clawdbot? i'm interested in contributing. i was thinking about it and multiplayer/multi tenant on single instance seems like it would be useful for org/family/"workspace" contexts

@adunne09 🫶🦞

clawdbot is incredible. I highly recommend trying it out. thanks @steipete clawd.bot

@UltraLinx immediately had to build this in a language learning context- mine is for spanish but you can easily edit it for another language. claude.ai/public/artifac…

Can you read 900 words per minute? Try it.

@adamdotdev what're your thoughts on the ghostty choice of not allowing users to create issues directly? github.com/ghostty-org/gh…

My first exposure to maintaining an open source repo is with OpenCode and I’m seeing firsthand the tidal wave of contributions that AI codegen has brought on. It’s a real problem and stresses me the fuck out lol, I hope we can collectively find some answers

all this talk about ralph im surprised windows hasn't been replaced yet

@rockorager nice! can you share the script/prompt? writing up some content rn to consolidate the various settings people use for ralph outside just the agent harness

One of the things we cache is svelte-check results because it's SO SLOW. svelte-check-rs is much faster, but also throws a ton of false positives on our codebase. Yesterday I started a rewrite. Did some design, started Ralph. Today, it's faster 28x faster than svelte-check, 1.73x faster than svelte-check-rs (and actually works on our codebase), 2.5x less RSS than svelte-check.