Live voice conversation
Speak through your microphone and hear spoken responses through your selected output device. Gemini Live handles the voice session.
THE FIELD GUIDE / 01 — Voice & two brains
Say it once. Let the right brain take it from there.
Live conversation for the moment. A task agent for the work that follows.
Jarvis, set a twenty-minute focus timer.
JARVIS / IN EVERYDAY LIFEIllustrated possibilities. Your tools do the work.
01 / THE POSSIBILITIESA little more
A little more
within reach.
You are in the middle of something. A timer should take a sentence; investigating a broken script needs a little more thought. Jarvis connects a live voice assistant to a separate task agent, so everyday desktop actions and deeper work have their own place in the same conversation.
See it in your dayTwo cooperating brains
Everyday function calls go to the desktop service. Research, writing and complex coding can be delegated to the high-level agent.
Interruption-aware audio
The audio path handles interruption and clears pending playback so the conversation can change direction. Actual responsiveness depends on your devices and connection.
Audio-file tools
The task agent can transcribe recordings, synthesize speech and play audio through configured providers. These are separate from the live microphone session.
02 / PUT IT INTO YOUR DAYOrdinary requests.
Ordinary requests.
Useful little transformations.
Follow the thought from the first sentence to the result. Make the names, files, and timing your own.
THE EVERYDAY / WORKFLOW 01
Start with something useful
You want to keep working without finding a timer app or typing a command.
START WITH THIS
“Jarvis, set a twenty-minute focus timer.”
- 01
Start with the context
The voice session receives your spoken request and selects the organizer function.
- 02
Let the work unfold
The desktop service stores a timer with a concrete duration and an identity you can refer to later.
- 03
Check the result
When it becomes due, the service records the event. Native alerts depend on Windows notification setup; voice announces results when the conversation is idle.
THE RESULT TO LOOK FOR
A saved timer that you can list, pause, resume or cancel. Starting it is separate from its eventual notification.
THE EVERYDAY / WORKFLOW 02
Hand a bigger question to the task agent
An explanation needs repository context and tool use, beyond an immediate desktop operation.
START WITH THIS
“Explain why the test in this project fails, inspect the relevant code, and suggest a small fix.”
- 01
Start with the context
Jarvis passes the requested task to the configured high-level agent through an authenticated local connection.
- 02
Let the work unfold
That agent can inspect files, search the project, plan and run permitted tools using its configured model.
- 03
Check the result
The returned result comes back into the voice conversation. Check the changed files and actual test output before treating the work as resolved.
THE RESULT TO LOOK FOR
An explanation and any permitted tool results. Delegation is not a promise that every project dependency or test environment is ready.
THE EVERYDAY / WORKFLOW 03
Turn a recording into a useful note
You have a recording on disk and want something you can return to later.
START WITH THIS
“Transcribe this audio file and write a short summary with the next steps.”
- 01
Start with the context
Supply or resolve the exact audio file so Jarvis uses the intended recording.
- 02
Let the work unfold
The configured transcription tool produces text; the task agent can organize that text into decisions and next steps.
- 03
Check the result
Ask it to save the summary to a named file, then open and review it. Speech synthesis can read the summary aloud if configured.
THE RESULT TO LOOK FOR
A transcript and, when requested, a saved summary. Unclear speech and speaker identity need human review.
Install voice and desktop dependencies
Run this in the extracted Jarvis source directory with uv installed.
uv sync --locked --extra voice --extra desktopConfigure both brains
Set GEMINI_API_KEY in the project-local .env. Configure the task model and its credentials with jarvis model. A local task model does not replace the cloud voice connection.
uv run jarvis modelChoose devices and start
List devices if the default microphone or speaker is wrong. The plain jarvis command also starts voice mode.
uv run jarvis lowlevel --list-devices
uv run jarvis run04 / A CLEAR PICTUREGood to know
Good to know
before you begin.
Voice needs a live connection
The default spoken conversation uses Gemini Live and an API key. Local storage and a local task model do not make the voice path offline.
An interrupted reply is not a rollback
Interrupting speech or timing out a delegation does not undo an action already performed. Inspect the task or activity result before retrying.
The natural next questions.
Workflow images are generated illustrations, not screenshots of a connected assistant. Availability depends on your configured tools and providers.
Context & project memory
Recognize the project, remember its other name, and keep the next step close.
Browser & workspaces
Organize your real tabs, carry useful information across apps, and return to a saved setup.
Files & media preparation
Find the file, read the screenshot, prepare a batch, and reuse the recipe.