Capabilities
Streaming agent execution with Claude CLI, step-by-step log, re-run
Persistent conversation with Jarvis across sessions
Gate destructive actions — ML + regex risk assessment
Regex patterns that bypass approval queue
Cron-based recurring task execution
Trigger → filter → step chains
Per-channel system prompt + model selection
Jarvis initiates contact on budget/metric/anomaly
Parameterized task templates
Multi-agent routing with isolated cwd + tool allowlist
Ephemeral execution with diff review before commit
14 channel adapters — Telegram, Slack, Discord, iMessage, Teams, Signal, WhatsApp, Matrix, Line, Mattermost, Google Chat, Twilio SMS/WhatsApp/Voice
Unified message log across all channels
4-tier trust (owner/trusted/readonly/blocked)
GitHub/Stripe/generic with HMAC verify
Fire events to external systems on 11 event types
18-plugin catalog with install/enable
Browse + install MCP servers — 10 curated
Cost tracking, model usage, task success trends
Daily/weekly/monthly cost caps with hard-stop
CPU/RAM/Disk time-series + cron tick log
5-min self-check with anomaly detection
Read-only SQL browser + query runner
Live screenshot stream from Mini with wake button
MCP tool registry with health status
Tail LaunchAgent logs
Public uptime page with 24h/7d/30d SLAs
Rotate channel tokens + webhook secrets
Unified timeline of every event
Self-audit of state changes
Unified notifications — approvals, pairing, anomalies, failures
OAuth hub: Gmail, Calendar, GitHub, Notion, Linear, Stripe
Multiple members with scoped permissions + API keys
Distributed node registry — multi-host Jarvis Hub
CSV/JSON download of tasks/messages/approvals/deliveries
Visualize workflow + channel routing
Generators + utilities (Slack manifest, etc.)
Auth + config + keyboard shortcuts
Sub-500ms bidirectional voice via Twilio Media Streams — needs Deepgram + TTS wiring
scripts/bench.js — regression tests for classifier + risk
Data model + API — UI drag-and-drop still TODO
File change preview for dry-run tasks — sandbox-based