shield
Privacy Policy

Privacy first.
Always.

JARVIS controls your digital life — but never at the cost of your data. Here's exactly how it works, in plain language.

calendar_todayLast updated Jan 2026 schedule4 min read verified_userGPLv3 · Open Source

Plain-language summary

JARVIS stores everything on your local disk. No central database, no analytics, no telemetry. Cloud LLM calls only send the minimum context needed for a request. Every irreversible action requires your explicit consent first. And because the entire project is GPLv3, you can audit every line of code yourself.

100% Local Storagelink

Your lifelong episodic memory graph, RAG document vectors (stored in ChromaDB), and rolling JSONL histories are saved exclusively on your local machine. JARVIS does not have a central database in the cloud. Your data never leaves your hard drive unless explicitly sent as a prompt to an LLM provider.

hard_drive
Local only
lock
Encrypted
database
No cloud DB
visibility_off
Not tracked

API & LLM usagelink

By default, JARVIS routes complex requests to cloud providers like Groq, Gemini, or Regolo. The data sent is limited only to the context required to fulfill your command. If you want true zero-knowledge privacy, you can configure JARVIS to use completely local models via Ollama or LM Studio by setting PRIMARY_BASE_URL and FALLBACK_BASE_URL in your .env file.

terminal

Fully local mode

Set both providers to http://localhost:11434/v1 (Ollama) or http://localhost:1234/v1 (LM Studio) for 100% offline inference with zero cloud calls.

Human-in-the-Loop (HITL)link

JARVIS connects deeply to your personal accounts (Gmail, WhatsApp, Telegram, Google Calendar). Background scout processes monitor these feeds securely — but JARVIS is strictly restricted by HITL protocols.

It will never send an email, alter a calendar event, or message a contact without explicitly asking for your permission first. Every irreversible action presents a consent gate with full context and a 90-second expiry window.

Zero telemetrylink

This project is fully open source. There are no hidden tracking scripts, no usage telemetry, and no analytics injected into the binaries or the Python source code. You are welcome to audit the entire codebase on our GitHub repository.

Contributor License Agreementlink

To protect the project's long-term sustainability, JARVIS requires all contributors to agree to our Contributor License Agreement (CLA). By submitting code, you grant the maintainer (Kaif Ansari) the right to use your contribution in both the open-source GPLv3 version and any future commercial licenses, ensuring the project remains fully funded and maintained without legal friction.

Your rightslink

Because everything runs locally, you have full control:

  • check_circle Delete any data — every memory, log, and vector lives in Data/jarvis_memory/ on your disk. Delete and it's gone forever.
  • check_circle Run fully offline — point both providers at local models and never send a single byte to the cloud.
  • check_circle Audit the code — every tool, prompt, and API call is public on GitHub under GPLv3.
  • check_circle Fork it — modify, self-host, and redistribute under GPLv3 with zero restrictions beyond the license.
  • check_circle Completely uninstall — one PowerShell command removes every trace: services, registry, PATH, and files. Automatic backup of .env and memory to Desktop first. See the uninstall guide.

Contactlink

Questions about this policy?

Open an issue on GitHub — I respond directly.

bug_report Open an issue