roshni 0.3.0

Personal data infrastructure framework — health, finance, journals, and AI agents

Roshni

roshni (روشنی) — "light" in Urdu

Safety-first personal AI assistant with Telegram and terminal support.

5-Minute First Run

You need Python 3.11+.

1. Install

pip install roshni

2. Run Setup Wizard

roshni init

The wizard guides you through:

• Bot identity (name + personality — with arrow-key selection)
• AI provider (Gemini default, with step-by-step setup guides for every provider)
• Bot's Brain (vault folder for memory, tasks, and knowledge)
• Integrations (Google services, Trello, Notion, Apple Health, Fitbit — each with setup walkthrough)
• Safety level (one simple choice: Balanced / Read only / Full access)
• Platform (Telegram or terminal)

3. Try It Immediately

roshni chat

Terminal chat is the fastest first experience. No Telegram required.

4. Run on Telegram

roshni run

Security Model

• Balanced (default): Can read and search everything, create drafts/tasks/notes. Asks before sending emails or deleting anything.
• Read only: Can search and read your data, but can't create or change anything.
• Full access: Can do everything including send emails and delete items.
• Telegram is deny-by-default unless your user ID is allowlisted.
• Gmail is draft-first by default (no send).

Integrations

Integration	Default	Risk	Notes
Gmail	Draft-only	Low	Saves drafts locally; optional send mode
Google Workspace profile	Off	Medium	Guided least-privilege scope choices
Obsidian	Off	Low	Read-only vault search
Trello	Off	Medium	Boards/lists/cards/labels/comments with approvals on write actions
Notion	Off	Medium	Database-backed page search/create/update/append
HealthKit import	Off	Low	Reads Apple Health export.xml summaries
Fitbit	Off	Low	Steps, sleep, heart rate via Fitbit API
Notes	On	Medium	Local note writes (approval-gated)
Weather	On	Low	Read-only
Web search/fetch	On	Low	Read-only
Reminders	On	Medium	Local reminder writes (approval-gated)

Local LLM (Ollama) Requirements

Recommended minimums:

• 8 GB RAM: small models only
• 16 GB RAM: solid everyday use
• 32 GB RAM: smoother larger local models

Quick start:

# Install Ollama first: https://ollama.com
ollama serve
ollama pull deepseek-r1

Then choose Local (Ollama) in roshni init.

Delighter Features

• morning_brief
• daily_plan
• weekly_review
• inbox_triage_bundle
• save_reminder / list_reminders / complete_reminder

Built-in Tools

• get_weather
• search_web
• fetch_webpage

Reconfigure Anytime

roshni init

Optional Extras

The base install includes everything you need: LLM, Telegram, Google APIs, web search. These extras add specialized features:

Extra	What it adds
health	Health data collection (pandas)
fitbit	Fitbit steps/sleep/heart rate
journal	Journal RAG with embeddings
financial	Financial calculators
financial-full	Market data + portfolio optimization
agent-langchain	LangChain integration
all	Everything above

Development

git clone https://github.com/uabbasi/roshni.git
cd roshni
uv sync --extra dev
uv run pytest tests/

See CLAUDE.md for architecture details.