nano

A FastHTML + MonsterUI web app starter for serverless deployments. Powers vedicreader.com.

Clone it, connect your blocks, ship it.

Getting started

git clone https://github.com/Karthik777/nano.git
cd nano
uv sync
uv run nano-setup       # link Vercel, scaffold .env.example, install skills
uv run python main.py   # http://localhost:5001

nano-setup is idempotent and safe to re-run. The console scripts shipped with the package:

script	purpose
uv run nano-setup	link Vercel, scaffold .env.example, install SKILL.md skills
uv run nano-skill	(re)install SKILL.md into .claude/skills/nano/ and .agents/skills/nano/
uv run nano-push	push .env values to the Vercel project (production + preview); --dry-run to preview

How it works

Each feature is a block: a self-contained module with its own config, routes, and database. You connect blocks to the app in order. Auth reads the full skip list at connect time, so it goes last.

# nano/app.py
b.connect(nano)   # blog
a.connect(nano)   # auth — always last

Each block exposes a connect(app) function that registers routes, seeds data, and wires up any middleware it needs. Blocks can share a database or borrow config from each other. They can also override routes registered by earlier blocks — first in line wins.

What's included

core handles config and the base UI (navbar, theme switcher, page layouts). Everything else builds on it.

auth covers email/password registration with Resend verification, Google OAuth, and GitHub OAuth. One connect() call sets up all routes and session middleware. Route paths are overridable via RouteOverrides.

blog is a full publishing block. Posts are seeded from Markdown files with YAML frontmatter. The list page uses a newspaper-style featured/sidebar/grid layout. Post detail pages support single-column or two-column newspaper layout, set per-post via layout: newspaper in the frontmatter. Code blocks never split across columns. To force a column break at a specific point in a post, add:

```col
```

Project structure

nano/
├── main.py
├── nano/
│   ├── app.py           # wire up blocks, scheduled jobs
│   ├── auth/            # auth block
│   ├── blog/            # blog block
│   └── core/            # config, UI
├── data/
│   └── db/              # SQLite databases
└── static/

Auth setup

Email/password:

RESEND_API_KEY=re_...

Google OAuth:

WANT_GOOGLE=true
GOOGLE_CLI=...
GOOGLE_SCRT=...
# callback: {DOMAIN}/a/google/callback

GitHub OAuth:

WANT_GIT=true
GIT_CLI=...
GIT_SCRT=...
# callback: {DOMAIN}/a/github/callback

Google and GitHub users are activated immediately. Email/password users get a verification link via Resend.

To change the default route paths:

from nano.core import RouteOverrides
RouteOverrides.lgn = "/login"
RouteOverrides.home = "/dashboard"
RouteOverrides.skip += ["/public"]

Extensions

The dev toolchain that ships with nano:

• kosha — indexes your repo and installed packages into a hybrid search + call graph database. Agents query it before writing code.
• fossick — get structured information from the web

Deployment

nano deploys through Vercel's native Git integration — push to main and Vercel builds.

1. Connect the GitHub repo to a Vercel project.
2. Attach Turso via the Vercel marketplace integration (provides TURSO_DATABASE_URL and TURSO_DATABASE_TURSO_AUTH_TOKEN).
3. uv run nano-push loads your .env into the Vercel project (production + preview). It leaves the Turso vars alone when an instance is already attached.
4. Push to main to deploy.

Local dev uses a SQLite file under data/db/; production persistence is Turso.

Style

No ruff, no PEP 8. The code uses fastai idioms: store_attr, patch, AttrDict, L. Short functions, no docstrings unless the function name isn't enough. It reads fine on a phone.

License

MIT