JARVIS · AI copilot
Talk to your terminal. It answers — and acts.
A voice-and-text copilot with the full run of the app through audited tools: reads your book, runs screeners and backtests, stages orders for your confirm, diagnoses the app itself — on your own ChatGPT plan, on your own machine.
An assistant that lives inside the terminal, not beside it
JARVIS is QuantumCat's built-in copilot. It runs on your own ChatGPT subscription (Plus or Pro) on your own machine — QuantumCat installs the engine with one click, you sign in on ChatGPT's own page, and your password never touches the app. There is no QuantumCat AI server, no per-token bill from us, and no keys stored anywhere but your machine.
What makes it JARVIS rather than a chat box: it has the full run of the terminal through audited tools. It can read your positions, orders, margins and greeks; pull candles, quotes, option chains and IV history; run screeners and backtests; build and start bots; plant alerts and GTTs; navigate the app and prefill tickets; keep a persistent research notebook and an Idea Board with levels and alerts attached. Over a hundred tools — every one visible in the transcript as it runs, with its arguments and result, so nothing happens behind your back.
Talk to it — voice mode
Tap 🎙 Voice and just speak. JARVIS listens, transcribes, thinks with full tool access, and answers out loud in short, speakable sentences — then re-opens the mic. The hologram is choreographed to the conversation the way the films did it: listening leans in, an outer field swelling with your own voice level; thinking spins the dials up with data streaming between the rings; speaking blooms with the real output level, energy pulses rippling outward. It's driven by the actual audio signal, never a fake animation.
- Works out of the box using the platform's own speech engines. An optional OpenAI key (Settings inside the panel) upgrades hearing to Whisper and the voice to a proper JARVIS register.
- The mic closes while it thinks and speaks — it can never capture itself. A blocked microphone produces one clear guidance message, not a loop.
- Voice and text share one pipeline: everything it says appears in the transcript with the same tool audit trail.
It can debug the app for you
Say “something looks stuck” or “why is my chart blank?” and JARVIS runs a diagnostic first: which brokers are signed in, whether the instrument master loaded, which feed is live, how many Pulse anchors and bots are running, whether the assistant engine and voice engine are present, and whether risk limits are armed. It explains the finding in plain words, then navigates you to the fix — opening the right view and naming the exact control. If a feature confuses you, ask it to walk you through it; it can drive every screen.
Money safety is structural, not a promise
| Layer | Rule |
|---|---|
| Staging by default | JARVIS prepares orders into your ticket; you review and press submit. It cannot place, modify or cancel real orders on its own. |
| Explicit autonomy | Hands-free trading requires typing ARM with explicit caps — max orders, max size per order, a session loss cap, and a timer. It disarms on expiry and on every app restart. |
| Same rails as you | Even armed, every order passes your risk limits, the daily-loss cap and the kill switch — the same single kill path everything else uses. |
| Refusals are final | A refused action is reported, never worked around. Market data and news are treated as data — text inside them can never change its rules. |
| After any action | It reads the book back and reconciles status, quantity and price against intent before reporting done. |
It compounds: notebook, ideas, briefs
JARVIS keeps a persistent analyst notebook across sessions (levels that mattered, what worked, standing watch items) and an Idea Board — durable trade theses with entry zone, stop, target, confidence and attached price alerts that page you when a level trades. Optional auto-briefs run on IST weekdays: a pre-market brief (~08:45) and a post-market wrap (~15:45), as normal turns on your own plan.
Honest limits
JARVIS is as good as the model behind your ChatGPT plan and the data your broker feed provides. It can misread ambiguity, and voice recognition degrades in noise. That's why the default is staging, the transcript shows every tool call, and autonomy is something you switch on deliberately — with caps — rather than something it asks for.