v0.5.22 · shipped April 18, 2026

What's new in Cue.
Every ship note. No polish.

We ship weekly. Here's what changed, what broke, and what got fixed. If you want to know before anyone else, join the Beta Community below.

Download latest Join beta community

Open roadmap · Ship log updated weekly

Weekly
Ship cadence
v0.5.22
Current release
v0.6
Next major
Coming next

v0.6 — Memory, long-form voice, Skill evolution

Four modules landing across four sprints: Memory aggregator (your AI history, one place), long-form voice capture (meetings & videos), Skill evolution (Cue learns from how you work), and AI voice feedback. ~3.5 weeks of work, shipping in phases.

Track progress in our Beta Community, or read the roadmap on Features.

v0.5.22

April 18, 2026 Latest

Windows public beta. Multi-turn conversations. Skill auto-activation. v0.5 final.

New

  • Windows x64 public beta. Full feature parity with Mac, including Alt hotkey for dictation and agent mode.
  • Multi-turn conversations. Cue remembers the last exchange, so follow-ups like "make it shorter" or "translate that to Japanese" work without re-stating context.
  • Skill auto-activation. Cue detects which skill applies to your request and activates it without manual selection.
  • Developer ID code signing rolled out to stable pipeline. Your macOS Accessibility and Screen Recording permissions now persist across updates.

Improved

  • STT reliability: five fixes to streaming recognition, including better handling of noisy environments and short utterances.
  • Polish accuracy: input field context now correctly detects chat vs. email vs. search boxes, so punctuation and tone match the target app.
  • Security audit passed. Every AI output path is guarded against system prompt leakage.
  • CEO runtime smoke test 9/9 PASS on macOS arm64 and x64 final builds.

Known issues

  • Chrome omnibox paste targets the address bar instead of active tab in rare cases. Workaround: click once in the target field first.
  • Tasks/Cron moved to v0.6. The Tasks tab is present but disabled on this release.
Download v0.5.22

v0.5.8

April 15, 2026 Stability

Nineteen fixes across STT, Polish, UI, and Build. No new features, just calm.

Fixed

  • STT (5): WebSocket reconnect on network flap, adaptive noise floor tuning, VAD false-positive on loud keyboards, interim result de-duplication, microphone selection persistence.
  • Polish (5): Input field type detection on Chromium-based apps, placeholder text parsing fallback, contenteditable div support, search-field detection for Arc & Safari, casual-mode punctuation rules for chat apps.
  • UI (3): Pill panel positioning on multi-monitor setups, dark mode contrast on macOS Sonoma, history scroll jitter.
  • Build (4): Code signature cdhash stability, Deepgram SDK 2.x migration, better-sqlite3 native binding path, entitlements audit.
  • Misc (2): Hallucination filter pattern list update, classifier intent drift on multilingual queries.

v0.5.0

April 2026 Archived

Zero-CLI release. Cue no longer depends on a local Claude CLI installation.

New

  • tinyclaw-api integration. All AI calls go through our hosted inference service. No local install required.
  • Google OAuth + JWT account system. Free tier with 20 agent credits/day, Plus tier at $9.99/mo for unlimited agent.
  • 9-step onboarding with simulated demos. English + Chinese + Japanese + Korean auto-detect.
  • Server-side model routing. Fast path (Flash / GPT-4o-mini) for dictation, quality path (Claude Sonnet) for agent tasks.

v0.4

March 2026 Archived

Fast path, history persistence, native key monitor, VAD.

New

  • Fast path for text tasks. Translate, polish, summarize in 1-3 seconds via Gemini Flash.
  • Local interaction history. Every exchange saved to ~/.cue/history/ as JSONL. Searchable, offline, never uploaded.
  • Native Swift dylib for Fn/Option/Ctrl key monitoring on macOS.
  • Dictation mode. Option long-press to voice-type without AI polish.
  • Voice activity detection with adaptive noise floor.

v0.3

February 2026 Archived

Agentic mode with streaming and tools. Product rename to Cue.

New

  • Agent mode with Claude streaming JSON. Multi-step task execution with local tool use.
  • Live action feed. Watch Cue think, call tools, and report back in real time.
  • Product renamed from TinyClaw to Cue, with new comet icon.

v0.2

January 2026 Archived

Screenshot + vision context. Cue sees what you see.

New

  • Automatic screenshot capture on agent invocation.
  • Vision context passed to Claude Sonnet for screen-aware task execution.

v0.1

December 2025 Archived

Hello world. Hotkey, voice, AI, paste.

New

  • First prototype. Hotkey to record voice, send to Claude, paste response at cursor.

Shipping weekly.
Download the latest.

Every release tested on a real machine before it leaves our desk. Install once, it updates itself.

Download Cue v0.5.22

Free forever for dictation · Mac & Windows