TypeWhisper 1.2 is out: faster startup, lower idle CPU, better AirPods/Bluetooth recovery, better short-clip handling, and a new Minimal indicator. Free, open source macOS dictation with local or cloud engines. https://t.co/JVoVN9eFLC
Introducing Claude Managed Agents: everything you need to build and deploy agents at scale.
It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days.
Now in public beta on the Claude Platform.
@fortelabs love this experiment. I've been building TypeWhisper around a very similar use case, voice-first work across apps instead of just dictating into one box. local by default, optional cloud if you want it, plus per-app and per-site profiles so the workflow can actually adapt.
nthropic employees get a completely different system prompt than external users. More detailed communication style. Basically a preview of what Numbat will bring to everyone.
The "be concise" prompt is a temporary patch until the model handles it natively.
I dug through Claude Code's full source code (~30MB, 1,902 TypeScript files). Found two internal model codenames and a bunch of things the public doesn't know. Thread:
All of this is from reading src/ only - no node_modules, no vendor code. ~96% of the 30MB is written by Anthropic themselves.
Source files: constants/prompts.ts, utils/undercover.ts, utils/modelCost.ts, buddy/types.ts, services/autoDream/
🎙️ TypeWhisper for iOS is now on TestFlight!
Local speech-to-text with a full keyboard extension. Your voice stays on your device — no cloud, no subscriptions, no BS.
Join the beta → https://t.co/tOPtWD5rzM
Feedback welcome! 🚀