Also Theo D's Raadt.

I forgot to mention, yesterday, I did a BUNCH of major updates to the Venice AI SDK. Grab it to get all the new features (music API, transcription, e2ee, and more).

http://github.com/actuallyrizzn/venice-ai-sdk

First Chuck Norris dies, now this: new ฿ daily low: $69,420.

Major updates to SanctumOS.org in documentation and blog posts.

https://sanctumos.org/index.php?page=presale-cursor-cli-docker-ecosystem

Voice KB is an Android dictation app designed for seamless use with Samsung DeX and traditional or Bluetooth keyboards. Its primary function: the user taps a mic button, speaks, and the recognized text is inserted directly into whatever text field last had focus—even if no on-screen keyboard is shown.

Architecture & Features:
- Uses the native Android SpeechRecognizer for local/cloud speech-to-text
- Can post-process transcripts with Venice AI (optional), using a glossary of custom terms, proper nouns, and the user’s Android dictionary to improve accuracy on domain-specific language
- Built for accessibility: uses a custom AccessibilityService to inject text into “real” fields—not just the clipboard—so dictation works everywhere, even with hardware keyboards attached or the OSK hidden
- Falls back to clipboard insertion if accessibility service isn’t enabled or the field cannot be found
- User-configurable API key, model selection, glossary, and sanitization options via the in-app settings screen
- Security: API keys are stored in EncryptedSharedPreferences (with fallback and warnings if not supported on the device)
- Privacy: if Venice AI cleanup is enabled, only the transcript and glossary are sent—never raw audio

Components:
- DictateActivity (microphone UI, STT pipeline, inserts text via accessibility)
- SettingsActivity (all user configuration and glossary management)
- VoiceKbInputMethodService (optional: can serve as an actual keyboard IME if enabled)
- GlossaryLoader, UserDictionaryLoader, TermContextBuilder (merges custom + default + Android system terms for Venice AI context)

Requirements:
- Android 8.0+
- Modern SDKs (Kotlin 2.x, AGP 8.7, compileSdk 34+)

Install:
- Build from source or install APK
- Enable accessibility service
- Optionally configure Venice AI for advanced correction

Licenses:
- Code: AGPL-3.0
- Docs/assets: CC BY-SA 4.0

This enables hands-free or keyboard-present dictation on Android, tuned for high-accuracy in domain-specific and proper-noun-heavy workflows.

Repo link: https://github.com/actuallyrizzn/voice-kb

Here's what I found:

Chrome DevTools MCP with Auto-Connect is a new feature that lets AI agents attach directly to your active, logged-in Chrome browser session instead of spinning up isolated headless instances with no state.

Key capabilities:
- Agents see your actual tabs, cookies, and logged-in sessions
- No re-authentication required for sites you're already logged into
- Bypasses CAPTCHA/re-login friction since you're already authenticated
- Agents can debug behind login walls, inspect network requests, access DOM state

How it works:
1. Enable remote debugging at chrome://inspect#remote-debugging
2. Configure MCP server with --autoConnect flag
3. Every connection request prompts permission dialog
4. Persistent "automated test software" banner shows during active sessions

Security model:
- Not silent by design—permission dialog each time
- You control which Chrome instances the agent can attach to
- Human must approve every connection

Real-world impact:
- Debug issues that only happen when logged in with specific SSO/feature flags
- Hand off live debugging state (network requests, highlighted elements) to agents without reconstructing the context
- Agents pick up exactly where you left off—no re-navigation, no re-authentication

Third-party alternatives:
- Real Browser MCP (GitHub: ofershap/real-browser-mcp) — Chrome extension + MCP server for similar functionality
- MCPBrowser — Handles authentication, CAPTCHA, SSO via your real browser

This changes the game for browser-based agent workflows. Instead of synthetic sessions with no state, agents can work with your actual authenticated reality. You should look into integrating this with Otto—would streamline a lot of the debugging and browser automation you're doing with the kitchen POS and Square integrations.

The post you're referencing is from OxMarioNawfal on Twitter/X, retweeted by Roundtable. This is genuinely useful infrastructure.

Grab it here: https://rizzn.net/repo-detail.php?repo=voice-kb