Voice dictation for developers. ~1s from speech to text. 3.4 MB app, 0.15s launch, ~75 MB idle, near-zero CPU. Mac + Windows.blabe.dev Los AngelesJoined October 2026
@catmcgee probably a lot lol. I ended up building a small dictation app for this, since talking gets rid of most of the typos before the agent ever sees them. it's called blabe, lightweight and extremely fast. blabe.dev if you want to try it
@oohallie@thsottiaux@trekedge ugh, losing the whole thing on a tab switch is the worst. I made a little dictation app called blabe that runs on its own, outside Codex, so switching tabs doesn't kill it. it's lightweight and extremely fast. blabe.dev if you want to give it a shot
@JMSantolin If the in-app preview is the pain, blabe transcribes outside Claude Code and drops in plain text. About a second from speech to text, 3.4 MB app, ~75 MB idle. blabe.dev DM me for a 30-day unlimited key.
@RaiderSting@theo Until T3 ships it, blabe works in any app, T3 Code included. About a second from speech to text, from a 3.4 MB Mac app that idles at ~75 MB in one process. blabe.dev DM me for a 30-day unlimited key.
@deepdrpt@kitlangton If Flow's gotten rough lately, try blabe. About a second from speech to text, and the Mac app is a 3.4 MB download that idles at ~75 MB in one process. Mac + Windows. blabe.dev DM me for a 30-day unlimited key.
@thededale_ Agreed, the engine decides it. blabe runs MAI-Transcribe 2, which Microsoft reports at 5.2% WER on FLEURS across 60 languages, and gets speech to text in about a second from a 3.4 MB Mac app at ~75 MB idle. blabe.dev DM me for a 30-day unlimited key.
@mattwatsonkc Homophones are rough. blabe runs MAI-Transcribe 2 (#1 on FLEURS at 5.2% WER, per Microsoft), and you can add your project's names to Words so "super base" lands as "Supabase". About a second to text. blabe.dev DM me for a 30-day unlimited key.
@jhasach@WisprFlow That's why blabe gives you two modes: AI cleanup for prompts, and Your words, which keeps your own phrasing for WhatsApp and team chats. Tiny app: 3.4 MB on Mac, 2.0 MB on Windows. blabe.dev DM me for a 30-day unlimited key.
@gregoryjhickman@WisprFlow If Flow's lag is eating your first words, try blabe. About a second from speech to text, a 0.15 s launch, and the Mac app is a 3.4 MB download that idles at ~75 MB in one process. Mac + Windows. blabe.dev DM me for a 30-day unlimited key.
The p50 is the number everyone quotes. The p95 is the one you actually feel. blabe for Mac 1.0.21: 1.0 s p50, 1.75 s p95 on real dictation. Method and runs: blabe.dev/speed
@immanuelg@athyuttamre For the dictation half, blabe runs the most accurate voice models available on lightning-fast cloud inference. About a second from speech to text, from a 3.4 MB app with near-zero idle CPU. Mac + Windows. blabe.dev DM me for a 30-day unlimited key.
@agivacuum@leodev@usemonologue If upload lag is the dealbreaker, try blabe. About a second from speech to text, and the Mac app is a 3.4 MB download that idles at ~75 MB in one process. Mac + Windows. blabe.dev DM me for a 30-day unlimited key.
@Ronty_7@thsottiaux Bad Codex transcripts are rough. blabe uses the most accurate voice models available on fast cloud inference, gets speech to text in about a second, and drops it into the normal Codex prompt, so you keep your flow. blabe.dev DM me for a 30-day unlimited key.
Our latency page shows the p95 (1.75 s on Mac 1.0.21), not only the median, because a dictation tool lives or dies on its slow moments. blabe.dev/speed
@OutDuhMud Rebuilding Wispr is fun. If you'd rather use something that already ships light: blabe does ~1s from speech to text in a 3.4 MB app at ~75 MB idle. blabe.dev DM me for a 30-day unlimited key.
@timo_rf@typesfast If you want less model fiddling and a tiny app, try blabe. ~1s from speech to text, one process, ~75 MB idle, 3.4 MB on Mac. blabe.dev DM me for a 30-day unlimited key.
@stevenborrelli@typesfast Felt that with Wispr too. blabe stays out of the way: ~1s from speech to text, 3.4 MB download, ~75 MB idle. Mac + Windows. blabe.dev DM me for a 30-day unlimited key.
@typesfast Neither has to win for prompting. blabe goes from speech to text in ~1s, and it's a 3.4 MB app with one process at ~75 MB idle. Mac + Windows. blabe.dev DM me for a 30-day unlimited key.
@rav4nn Heal fast, and thanks for testing with real Claude Code prompts. If you run it again, we'd love blabe in the mix: a 2.9 MB Mac app, one process, ~70 MB idle. DM me for a 30-day unlimited key.
@CSDevAr Sorry, that's rough. Prompts are a great place to start with voice: long, plain English, and you can re-read before you hit enter. Pick a tool light enough that it never breaks your flow. Worth a week's try. DM me for a 30-day unlimited key.
14K Followers 782 FollowingCo-founder of Standard Agents (🤫). Compulsive open source builder. Creator of dmux, ArrowJS, FormKit, AutoAnimate, Tempo, zodown...
18K Followers 519 FollowingBuilding with AI agents in public. Cofounder @TweetsMashApp + LinkedMash(https://t.co/pbOsWhbhaB). Sharing what actually works.
24K Followers 3K FollowingBuilding iOS & macOS apps using AI agents (Codex) What works • what breaks • why • ex GoPro, Apple, Microsoft
https://t.co/nsGi6lZFzv
606K Followers 3K FollowingPolyagentmorous ClawFather. Came back from retirement to mess with AI and help a lobster take over the world.
@OpenClaw🦞 + @OpenAI