run small software
your agent wrote it.
now it runs.
deploy small software in seconds. share it with your team like a doc.
hand this to your agent
$ set up https://tinyrun.sh/SKILL.md
and let it take it from there. it installs the CLI, reads the contract, and deploys without asking you how.
Running tinyrun deploy ./budget: packs the folder, detects Python 3.12 with Streamlit, builds the image, starts the container, passes the health check, routes the subdomain, and reports the app live at budget.acme.tinyrun.sh in 4.2 seconds.
that's what your agent runs. one line.
HOW IT ACTUALLY WORKS
claude writes a folder of code, or you do
tinyrun deploy ./app
we build it, host it, and give it a URL on https
→ app.tinyrun.sh
your teammate clicks the link, gets a magic email, is in
no signup, no password
the app sleeps when nobody's on it, wakes in about a second
sleeping = $0
that's the whole wake screen. it's gone before you've read it.
export your code + sqlite anytime. your apps are yours.
WHAT PEOPLE BUILD
all eight are live right now. click one, add a row, come back tomorrow. it'll still be there.
FOR TEAMS
incident log
what broke, when, and what fixed it. still-open ones stay at the top.
FOR PEOPLE
shared shopping list
flatmates tick things off. whoever's at the shop sees what's left.
agents made software easy to write.
nothing made it easy to share.
- ·most software only ever needs a handful of users. that's a feature, not a failure.
- ·the person deploying is no longer a developer. tooling should stop assuming otherwise.
- ·sleeping software should cost nothing. you pay for what runs.
- ·your code, your data. small software deserves the same ownership as big software.
THE OBVIOUS QUESTION
why not just cloudflare or vercel?
fair question. if pages or vercel covers your case, use them — they're excellent. here's where the line actually is.
they host sites. this hosts programs.
pages and edge functions have no long-lived process, no filesystem, no sqlite file sitting next to your code. the moment your agent writes a python server with a database, someone is rewriting it into workers + d1 + kv + r2 and a bindings config. here the folder runs as written: container, https url, persistent volume, cron. one command.
their errors are written for you. ours are written for your agent.
a failed deploy comes back as a structured payload — stage, logs, healthcheck result — so the agent fixes its own mistake instead of pasting a screenshot at you. deploys are health-gated and atomic: a broken build never replaces the version that's serving.
their dashboards are for traffic. your tools have three users.
waf rules, bot scores and analytics are built for public sites. most agent-written tools are internal — a handful of people behind an auth wall, and nobody else should ever see it. that wall is already up here, per app, before your first deploy.
and if you outgrow us: export your code + sqlite anytime and leave. no lock-in is the feature.
common questions.
anything that's a folder of code: python, node, static sites, streamlit, fastapi. if it runs locally, it runs here. docker optional, we detect most stacks automatically.
no. point your agent (claude, cursor) at tinyrun's SKILL.md and say "deploy this to tinyrun." the agent runs the CLI, you approve. the CLI is there when you want it.
you decide, per app. a private app needs a tinyrun sign-in. an anonymous visitor gets bounced to the login page and never reaches your app, and the container never even wakes, so nobody can burn your CPU knocking on a door they can't open. want tighter than that? name the exact email addresses on the app's page and everyone else is out, including the rest of your workspace. the apps above are public on purpose so you can click them.
nothing. zero. an idle app stops, and the next request starts it again. we stormed our own box with twelve cold starts at once and the slowest took 0.29 seconds. you pay for the seconds your app spends actually serving someone.
yes. export your full source + sqlite file anytime, one command. apps are standard containers, so if we disappeared tomorrow you'd have everything.
they're built for production software with traffic. tinyrun is built for tools with 3 users: no yaml, no build config, no billing surprise, and auth is already done. the longer answer is a few scrolls up.
start deploying
no waitlist, no card. install the CLI and sign in with your email.
$ curl -fsSL tinyrun.sh/install | sh