OpenYabby

First public release of OpenYabby.

Added

• Voice pipeline — WebRTC bidirectional audio with the OpenAI Realtime API. Wake word (Yabby) detection via Silero VAD + Whisper transcription. Optional speaker verification via SpeechBrain ECAPA-TDNN microservice (npm run speaker).
• Multi-agent orchestration — Hierarchical lead / manager / sub-agent system with phase-aware prompts (discovery, planning, execution, review, QA). Auto-trigger of parent reviews on sub-agent completion.
• Multi-agent task cascades — Position-based parallel/sequential execution across agents, with one-shot in-memory completion bus replacing DB polling.
• CLI runner registry — Five runners supported out of the box: Claude Code (default), OpenAI Codex, Aider, Goose, Cline, Continue. Per-runner native session id persisted for correct resume.
• Channels — Discord, Slack, Telegram, WhatsApp, Signal. One-owner-per-channel pairing model. Unified LLM function-calling loop across voice and channels.
• 37 connectors — Static catalog with built-in and MCP backends. Tools auto-prefixed conn_* / mcp_* for namespace isolation.
• Persistent memory — Mem0 + Qdrant (file-based) for facts extracted every 6 turns. Hybrid search (vector + keyword), temporal decay, query expansion.
• Plan review and discovery — Lead agents submit plans for user approval and ask discovery questions through SSE-driven modals (voice, modal, or connector form types).
• Scheduler — In-memory tick loop supporting interval, cron, and manual schedules with retry and orphan recovery.
• Relay tunnel — Optional WebSocket tunnel to relay.openyabby.com for remote access. Auto-reconnect with exponential backoff. Disabled by default.
• Plugin system — Drop-in plugin loader with scoped logger, tool registry, event bus, and HTTP route mounting.
• Auth gateway — Optional username/password protection with API tokens and session cookies. Exempt paths for the SDP and SSE endpoints.
• Multi-channel media store — Content-addressed (SHA-256) media assets with PDF/vision/audio extraction.
• 41 idempotent migrations — All run on startup, listed explicitly in server.js startup().
• i18n — French (default), English, Spanish, German via lib/i18n.js. Frontend locales in public/locales/.

Known limitations

• macOS-first (Linux/Windows on the roadmap)
• Voice pipeline requires the OpenAI Realtime API (local-voice path planned)
• Claude CLI required for the default runner (other runners available)
• WhatsApp adapter pulls in @whiskeysockets/baileys which transitively includes a GPL-3.0 component — see THIRD_PARTY_LICENSES.md

Install

git clone https://github.com/OpenYabby/OpenYabby.git
cd OpenYabby
./setup.sh

See the README for full setup, configuration, and architecture documentation.