Always-ready voice
openWakeWord · faster-whisper · TTS
Jarvis.
A hands-free assistant that listens, understands, routes, and acts—with deterministic speed before LLM reasoning.
ListenRouteToolsMemorySystem
Wake word detected
At your command.
Intent routerTier 0 / Fast
ExecutionConfirmed
USER Start bug bounty mode
JARVIS Workflow active
JARVIS Workflow active
01 / Product
A voice interface for the whole machine.
Jarvis turns natural speech into fast, controlled action across media, browsers, apps, and Windows.
Say “Hey Jarvis,” speak naturally, and the system moves from wake-word detection through speech recognition to the right execution tool.
Common phrases bypass the LLM entirely. Vocabulary, heuristics, search context, and Ollama layers engage only as the command becomes more ambiguous.
233Stress-test phrases
03Routing tiers
LocalSTT + LLM runtime
02 / Capabilities
Hear it. Route it. Execute.
Layered routing
Fast path · vocabulary · heuristics · Ollama
Media control
Playback · volume · tracks · fullscreen
Browser + system
Tabs · search · apps · lock · restart · shutdown
Workflows
Study · assignment · bug bounty · music
03 / Routing
Fast first. Smart when needed.
Deterministic path
Exact commands and familiar phrases route instantly without invoking an LLM.
Context layer
Vocabulary correction, music entities, search context, and heuristics clarify intent.
Reason + execute
Qwen resolves ambiguity against a narrowed tool schema, then Jarvis performs the action.
04 / Stack
Built like a system. Feels like magic.
PythonopenWakeWordfaster-whisperOllamaQwen 3.5PlaywrightMusicBrainzWikidataWindows
05 / Get the product
Bring your own Jarvis.
$ git clone https://github.com/mahirhacks/Jarvis-Voice-Assistant.git
View repository ↗