"Part personal assistant. Part coach. Part brutally honest best friend who texts you at 9am to ask why you didn't go to the gym yet."
Jarvis isn't something you open. It's something that runs in the background, reading your calendar, watching your habits, pinging your phone when you're slipping, and waiting in the terminal or web app when you want to talk.
Push notification with today's calendar, the weather, your top 3 commitments, and one sentence on what actually matters today.
Habit slip detected. Push notification: "Gym was on your list. It's been 4 days. What's the actual reason?" And yeah, it follows up.
Meeting with Julie in 30 min. Jarvis surfaces your last 3 conversations with her, the deliverable you owe her, and her birthday next week.
Voice memo from your phone → transcribed, parsed into facts/commitments/people, indexed semantically. No app to open.
"You committed to 3 things this morning. Two are done. What's the story with the third? Real obstacle or excuse?"
Mood + energy logged, journal prompt, tomorrow's top 1 set. Sleep with the system off. Quiet hours kick in until 7am.
Push notifications are the always-on channel. Jarvis pings you the way a real friend would: context-aware, timed for when you'll actually act, never spammy. Quiet hours, escalation logic, and a one-push rule built in.
Miss a habit? You'll hear about it. Got a meeting in 30? Here's everything you need. Ghosted a commitment for 3 days? It's going to say something.
Every morning at 7am, Jarvis pulls your Google Calendar, the weather, your active commitments, and the people you have meetings with, and gives you a single coherent picture of the day.
Not a wall of bullet points. Not 14 notifications. One brief. One direction.
Most AI tools are stateless. They forget you the moment a conversation ends. They tell you what you want to hear. They give you bullet points when you need a friend.
Jarvis is built around a local knowledge graph and a tiered context engine. Every conversation, fact, person, goal, and emotion gets stored, indexed, and made retrievable, so the AI actually grows up with you.
Token-budgeted retrieval across identity, accountability, people, and semantic vector search powered by ChromaDB. The right context every turn, never the whole brain.
Built on CBT, Motivational Interviewing, Self-Determination Theory, Wheel of Life, Implementation Intentions, and Decision Quality. Not pop self-help. Actual frameworks.
Distress signals override every other directive. Routes to 988 and Crisis Text Line, drops the coaching mode, prioritizes a human handoff. Safety before everything else.
Every memory, every fact, every conversation lives in JSON files on your machine. No cloud sync. No training data harvested. Yours.
The accountability engine isn't gentle. Every commitment has a deadline. Every excuse gets logged. Every pattern of avoidance gets named. If you've blown off the gym four times this week, you're going to hear about it, by name, with receipts.
Earned praise only. No "I'm so proud of you!" for brushing your teeth.
Every person you mention gets a profile. Every fact you share gets stored, timestamped, and indexed. Every goal, every habit, every emotional pattern. All of it queryable, all of it surfaced when relevant.
You never re-explain yourself. Jarvis already knows.
The system prompt explicitly bans menus, sycophancy, preamble, and service-offer endings. Jarvis talks like a sharp, honest friend who already knows the situation.
Terminal, web, phone, voice. Same memory, same personality, same standards. Whether you're typing on a laptop at 1am or dictating into your phone walking to class, you're talking to the same person.
And that person has been paying attention.
# Tuesday, 9:47pm you > kinda dreading the meeting w/ Julie tomorrow jarvis > Because of what she said Friday or the deliverable you still haven't sent her? You logged it as a commitment 3 days ago. Send the draft tonight, then the meeting becomes a normal conversation instead of a defense. you > yeah fair. ok pulling it up now jarvis > Good. Marking it as in-progress. I'll check in at 11.
Python core, multi-LLM backend, local persistence, and a deliberately simple deployment model. One machine, your data, no servers to pay for.
The end state isn't an app on your phone. It's a presence in your home. A standalone device that sits on your desk or counter, always listening, always remembering, no screen required.
Say "Hey Jarvis" from across the room. Whisper STT runs locally on the device. Sub-second response time, no cloud round-trip.
The device is the agent. Microphone array, far-field listening, soft ambient speaker. Your phone becomes optional, not central.
Identical memory, identical personality, identical accountability. Walk into your kitchen at 7am and get the morning brief out loud.
Local-first stays local-first. Voice processing on-device, memory on-device, your home stays your home. Nothing leaves unless you say so.
Foundation is solid and in daily use. The road ahead is hardening, polishing, and the final integrations before public beta.
Knowledge graph, commitment tracking, excuse detection, tiered context assembly, vector search.
CLI, web dashboard, push notifications, scheduler, proactive nudge engine with quiet-hours.
CBT pattern detection, Wheel-of-Life scoring, decision journaling, intervention tracking.
Whisper STT and TTS wired. Currently tuning the conversational latency loop and personality match.
Mobile-first dashboard, commitment toggles, life-score visualizations, conversation timeline.
Read/draft email triage, calendar conflict resolution, meeting prep auto-context.
Add full test suite, error degradation paths, async ChromaDB calls, and conversation log scaling.
Onboarding wizard, packaged installer, opt-in encrypted backup, documentation. Coming soon.
Across 29 modules. 60% memory and accountability, 25% interfaces, 15% integrations.
Each backed by structured JSON with timestamps, supersession tracking, and conflict detection.
Morning briefing, evening review, commitment checks, habit nudges, weekly reviews, relationship pings.