JARVISYour personal intelligence.
J
JARVIS
What are we working on?
Ask a question, explore an idea, or give me something to do.
JARVIS
Workspaceworkspace // open canvas

Run tool

Execution uses the indexed, integrity-checked source.
Layouts

Workspaces

Capture current layout
Loading workspaces…
Library

Interfaces

Import a skill

Paste a local Python implementation and define when JARVIS should call it.

Capabilities

Verification: selecting Import validates syntax, imports, context adapters, capabilities, complexity, and the exact entry-point before writing anything. Arbitrary network, system, hardware, secret, shell, dynamic-code, and direct file APIs are rejected.

Imported code is SHA-256 indexed and will not execute if modified.

Import an interface

Add a local HTML/CSS/JS folder as a saved, reusable interface.

No folder selected

Verification: the folder is wrapped in the same sandboxed, strict-CSP viewer as generated interfaces. Network access, storage, eval, and external resources are rejected; local <link>/<script src> references inside index.html are inlined automatically.

Only .html, .css, and .js files are read.

Groq triple core

One fast provider, with separate chat, tool-making, and interface roles.

Security: use a newly rotated key. When remembered, the key is stored in macOS Keychain—not project files or browser storage—and restored when JARVIS restarts. macOS may ask you to approve Keychain access.

All three roles are checked against Groq’s live model list.

Exa

Connect Exa for AI-native web search.

Security: the key is used only by the fixed api.exa.ai adapter. It is never returned to this page, exposed to generated tools, or sent to either AI model.

The live check uses one search request. Local SearXNG remains available automatically if you disconnect Exa.

Exa is preferred over local search whenever it's connected.

OpenWeatherMap

Connect the Current Weather and Geocoding APIs.

Security: the key is used only by the fixed api.openweathermap.org adapter. It is never returned to this page, exposed to generated tools, or sent to either AI model.

The live check uses one current-weather request. New OpenWeatherMap keys can take a short time to activate.

Your exact GPS is requested separately by the browser.

Settings

Connections, credentials, and behavior

Camera vision

Let JARVIS see through your webcam. While enabled, sending a message shares a snapshot with your connected Groq vision model. No audio or continuous video is sent; images are not saved in chat history.

Camera off

Model core

Offline
Chat // Groqopenai/gpt-oss-120b Tool maker // Groqopenai/gpt-oss-120b Interface maker // Groqopenai/gpt-oss-120b Credential not configured

Local model

Not installed
Qwen3-4B-Instruct-2507-Q4_K_M // llama.cpp (Metal) Runs fully offline on this machine. Noticeably smaller and less reliable than Groq's default model for building tools and interfaces — good for quick chat, not a drop-in replacement.
llama.cpp runtime — not installed (~11 MB from GitHub)
Qwen3-4B (Q4_K_M) weights — not installed (~2.5 GB from Hugging Face)

Web research

Offline
Search // Local SearXNG127.0.0.1:8888 No API key · live results · source links

Live weather

Offline
Conditions // OpenWeatherMapcredential required Current Weather API · precise GPS supported

Mobile inbox

Offline
Pairing // JARVIS Inbox Relay No phone paired yet

Voice

Output off
Speech // Microsoft Edge neural TTS

Wake word

Off
Hands-free // say “JARVIS” to start talking While on, your mic stays open and browser speech recognition listens continuously for the word “JARVIS.” Saying it starts capturing your message; JARVIS sends it automatically once you stop talking. Audio is processed by your browser's speech service, same as the mic button.

Backup

Not synced yet
Mirror // everything you’ve saved Tools, interfaces, workspaces, notes, and everything else you’ve saved are mirrored here automatically — not JARVIS’s own app code.

Shape JARVIS’s soul

Feedback persists

Teach JARVIS how to communicate and prioritize. Immutable safety principles cannot be removed.

AI areas

Assign an AI to each area of work, and keep a reusable prompt library for it.