ryu-app

Everything happening on this node (monitor alerts, finished tasks, runs, approvals, meetings, notes), newest first, grouped by day.

A persisted, named, ordered collection of agents plus a coordination strategy, addressed as one unit via @team.

A Windows-only experimental system-wide predictive-typing companion (Win32/UIA): inline ghost text in any text field, accepted with Tab. Compile-only, unshipped; mac/linux planned.

Every agent and workflow scheduled job on Month/Week/Day/Agenda views (past runs from history, upcoming from cron/interval) with a New-automation dialog.

The universal TTS sidecar: a self-contained Python HTTP front over several text-to-speech engines (Kokoro, Kitten, Pocket), part of Ryu's speech data path.

Price / stock / keyword / content-diff / uptime watches with cross-device notification fan-out.

Agent-native Loom/Jam: capture and browse screen/timeline clips via the Shadow capture proxy.

A customizable, constantly-updating grid of widgets.

A human-in-the-loop queue where agent-proposed actions, edits, and fixes wait for your accept or reject before they take effect.

Petgraph DAG automation with triggers, durable execution, and a natural-language workflow builder.

A ComfyUI / ElevenLabs-Flows-style node board: wire up image / video / chat / tts / stt / upload / note nodes through Ryu's media and agent bridges; each board persists as a Space document.

Record a call, transcribe it live, and generate structured AI notes (Granola/Notion-AI style). (Python diarization sidecar held back, not yet public.)

Drive iOS Simulators (simctl, macOS + Xcode) and Android Emulators (adb) from a workspace tab: list/boot devices, install & launch apps, open deep links, screenshot, and (Android) tap/swipe/type, exposed as the grant-gated simulator.control capability.

Author a user-owned Agent Skill (SKILL.md): front-matter fields, a markdown body editor, debounced autosave, and server-backed undoable version history.

A LoRA/QLoRA training studio: durable job state, the trained-adapter catalog, and the Python training sidecar.

Parameterized, replayable native-desktop automations: record once with a frontier model, replay forever with a small one.

Failed runs are diagnosed (Gateway side-model) and a fix is proposed to the approvals inbox.

An auto-detecting todo list surfaced from your chats and activity, tracked as a lightweight quest board.

An Excalidraw whiteboard shipped as a Ryu app; draw freely, each board persists as a Space document, with Mermaid-to-Excalidraw diagram import.

A CapCut-style scrubber over Shadow's captured screen/input/window/audio lanes with keyframe preview, a Dayflow work journal, and jump-to-moment focus.