Jarvin

Jarvin is a private, host-run personal AI assistant.

It is designed around one trusted machine doing the real work:

• local LLM inference
• voice processing
• persistence
• tool execution
• integrations

Desktop and phone clients then connect to that host over local network or WireGuard.

Current Capabilities

• local llama.cpp runtime with optional headless Ollama backend
• typed chat with multi-conversation history
• shared React client served at /app/
• Tauri desktop app
• Tauri Android shell
• remote phone voice:
  • phone microphone capture
  • host-side transcription
  • phone speaker playback of Jarvin replies
• SQLite-backed profile and conversation memory
• reminders and routines
• morning briefs that combine weather, calendar, and reminders
• Google Calendar integration with event CRUD
• DuckDuckGo-backed web research with page fetch and summarization
• safe host-side assistant tools for workspace and repo tasks
• natural-language planners for:
  • weather
  • calendar
  • reminders
  • workspace actions
  • research
  • daily briefs

Product Shape

Jarvin is not trying to become a custom foundation model.

The project is focused on building a strong assistant system around local models:

• voice input and output
• memory
• tools
• integrations
• clients
• planning and follow-up handling
• proactive assistant behavior over time

Quick Start

Create the virtual environment:

py -3.11 -m venv .venv

Install dependencies:

.\.venv\Scripts\python -m pip install -r requirements-gpu-cu128.txt
.\.venv\Scripts\python -m pip install -e .

Run the host:

python server.py

Then open one of:

• http://127.0.0.1:8000/app/

Clients

Tauri Desktop

cd clients\jarvin-ui
npm install
npm run tauri dev

Host-Served Shared Shell

cd clients\jarvin-ui
npm run build:host
python server.py

Then open:

http://<host-or-wireguard-ip>:8000/app/

Tauri Android

cd clients\jarvin-ui
npm run tauri:android:pixel:debug

Primary APK artifact:

clients/jarvin-ui/artifacts/jarvin-mobile-arm64-debug.apk

Example Prompts

Weather:

What's the weather in Burnaby near Metrotown?
How about tomorrow?

Calendar:

What's on my calendar today?
Put lunch with Sam on my calendar tomorrow at noon.
Move lunch with Sam back an hour.
Delete lunch with Sam from my calendar.

Reminders and briefs:

Remind me to call mom tomorrow at 5pm.
Every weekday at 8am remind me to stretch.
Give me my morning brief.

Workspace and research:

Could you look through the codebase for include_router?
Pull up backend/api/app.py lines 10 to 30.
Research llama.cpp windows cuda docs for me.
What else did you find?

Integrations

Search

Default search provider:

• DuckDuckGo Lite

Validate search:

.\.venv\Scripts\python scripts\validate_integrations.py --search-only --query "llama.cpp windows cuda docs"

Google Calendar

Save your OAuth desktop client JSON at:

secrets/google-calendar-client.json

Then ask Jarvin:

connect my Google Calendar

Validate calendar setup:

.\.venv\Scripts\python scripts\validate_integrations.py --calendar-only

Repo Guide

Important paths:

• server.py: host entrypoint
• config.py: settings and .env loading
• backend/api/app.py: FastAPI app assembly
• backend/api/routes/chat.py: chat endpoint
• backend/agent/chat_tools.py: planner and tool router
• memory/conversation.py: conversations and profile
• memory/reminders.py: reminders and routines
• clients/jarvin-ui: shared desktop/mobile client
• docs/runbook.md: operational instructions
• docs/architecture.md: current system shape
• docs/roadmap.md: product direction and next phases

Testing

Backend suite:

.\.venv\Scripts\python -m pytest tests\backend -q

Integration validation:

.\.venv\Scripts\python scripts\validate_integrations.py --search-only
.\.venv\Scripts\python scripts\validate_integrations.py --calendar-only

GPU diagnostics:

.\.venv\Scripts\python scripts\diagnose_gpu.py

Detailed Docs

• Runbook
• Architecture
• Roadmap
• Client README