TinyKit – Self-Hosted Agentic App Builder

TL;DR

TL;DR: TinyKit is an open-source, self-hosted platform that lets you build, tweak, and deploy small web apps using an AI agent — all from a single server powered by PocketBase.

Source and Accuracy Notes

• Official site: https://tinykit.studio
• GitHub: https://github.com/tinykit (star the project)
• HN Launch: Show HN: Tinykit – self-hosted Lovable, deploys to itself

What Is TinyKit?

TinyKit is an open-source agentic app builder in the same vein as Lovable — you prompt an AI agent to build a web app, and it writes the code, creates the backend, and wires up the UI. The key difference is that TinyKit is fully self-hosted. Everything runs on a server you control, powered by PocketBase. When the app is ready, it exports as a static HTML bundle that loads near-instantly.

The workflow is designed for small, focused apps rather than large products. You start from a kit (starter template) or build from scratch. The agent handles the coding; you handle the prompting and visual tweaking. Each app gets its own JSON data collections with realtime sync, built-in image storage, and a visual design system editor that updates CSS variables without touching code.

Apps are compiled to static HTML at deploy time. The result is a single-file Svelte app that loads fast and can run anywhere, including on a shared server alongside dozens of other TinyKit apps.

Setup Workflow

Step 1: Install on Your Server

TinyKit runs on any server that can host a Node.js app. The recommended path is a VPS with Docker or a manual Node install.

# Clone the repo
git clone https://github.com/tinykit/tinykit.git
cd tinykit

# Run the setup script
./setup.sh

The setup script installs PocketBase, configures the database, and starts the studio UI.

Step 2: Connect Your LLM Provider

TinyKit supports Claude, GPT, and Gemini. You bring your own API key and see token costs in real dollar terms per prompt.

# In the studio, go to Settings → AI Providers
# Add your API key for your preferred provider
# TinyKit shows cost estimates before each agent run

Step 3: Create Your First App

From the studio dashboard, click New App. Choose a kit (starter template) or start blank. Describe what you want in plain language — the agent writes the code, creates the data model, and sets up the routes.

Prompt example: "Build a simple habit tracker with daily check-ins, 
a streak counter, and a weekly summary chart."

The agent works in a loop: write code → show preview → you request changes → agent revises. You can also edit the generated Svelte file directly if you prefer hand-coding.

Step 4: Deploy

When you’re happy with the app, hit Deploy. TinyKit compiles it to a static HTML bundle and serves it from your server. You get a URL like https://yourserver.com/app-name.

# Or deploy via CLI
tinykit deploy --app my-habit-tracker

Deeper Analysis

Architecture

TinyKit has two layers: the studio (where you build) and the runtime (where apps live). The studio is a web UI backed by PocketBase. Apps are static Svelte bundles that connect to PocketBase for data. This means every app shares the same backend infrastructure but stays isolated in its own static deployment.

The AI agent is called on-demand during the editing session. It has access to the app’s current file, the data schema, and the design tokens. It does not have persistent memory across sessions — each build/edit cycle starts fresh.

Starter Kits

TinyKit ships with several official kits:

| Kit | Purpose | |---|---| | launchkit | Landing pages and waitlist signups | | bizkit | Small business: invoices, expenses, CRM | | devkit | Developer: docs, changelog, status page | | dailykit | Personal: notes, habits, bookmarks, recipes | | freelancekit | Solo business: invoices, time tracking, proposals | | creatorkit | Audience building: link in bio, newsletter, analytics | | sitekit | General web: homepage, blog, contact form | | agencykit | Agency: projects, clients, team portfolio |

Community kits are on the roadmap.

AI Cost Tracking

One practical detail: TinyKit shows you the exact dollar cost of each agent run before you approve it. This is useful for keeping LLM spend under control, especially when vibe-coding means multiple revision cycles per app.

Realtime Data

JSON collections are created per app and sync in realtime via PocketBase subscriptions. For a habit tracker or a notes app, this means live updates across tabs without polling.

Practical Evaluation Checklist

• [ ] Server with Docker or Node 18+ available
• [ ] PocketBase runs without conflicts on your port setup
• [ ] LLM API key ready (Claude/GPT/Gemini)
• [ ] First app builds and deploys in under 10 minutes
• [ ] Static bundle loads fast on your target device
• [ ] Token cost estimates match your expectations

Security Notes

• All data stays on your server. No third-party data sharing.
• PocketBase handles auth for the studio; each app can have its own auth rules.
• API keys for LLM providers are stored in the studio settings, not in app code.
• Static deploys mean no server-side code execution for deployed apps — attack surface is the static file server.

FAQ

Q: How is TinyKit different from Lovable? A: The core difference is hosting. Lovable is a SaaS product where your apps live on their infrastructure. TinyKit is fully self-hosted — you run the studio and the runtime on servers you control. If you want to own your app platform and avoid SaaS lock-in, TinyKit is the self-hosted alternative.

Q: Can I use TinyKit without AI? A: Yes. You can hand-code directly in the built-in code editor. The agent is an optional co-pilot, not a requirement. The design system editor and realtime data collections work without any AI involvement.

Q: How many apps can I run on one server? A: The marketing says 100+ apps on a single server. Realistically, the number depends on traffic and app complexity. Each app is a static bundle with a PocketBase connection, so the server load is light unless you have high traffic.

Q: Does it work on mobile? A: Yes. The studio is responsive — you can prompt the agent, tweak settings, and deploy from a phone browser.

Conclusion

TinyKit fills a specific niche: self-hosted, AI-assisted app building for small, focused tools. If you want to offer a “build an app” capability to a team or community without relying on a SaaS platform, it is a compelling option. The starter kits cover common use cases, the static deploys keep performance high, and the BYO LLM model means you control your AI spend. The main trade-off is that you are managing the infrastructure yourself — if you want a fully managed experience, Lovable or similar SaaS tools are still the easier path.

For developers comfortable with self-hosting who want an agentic app builder they fully own, TinyKit is worth a look.