The archive · Developer & Business Tools · Product decision · 2023–2026
Dropbase bets AI-written, editable Python apps beat no-code; 1.3k GitHub stars
Prompt-to-app builder whose AI writes Python on its own framework, local-first and self-hosted; HN launches Dec 2023 and Jul 2024, 1.3k stars by Sep 2026.
Dropbase
What the business is
Dropbase is an AI-based Python web app builder for admin panels, back-office tools and internal dashboards: a developer describes the app in natural language and the model writes Python code on Dropbase's own web framework with prebuilt UI components; changes appear in a diff the developer can verify and edit before previewing, all running locally with the developer's own LLM keys.
How it started
Dropbase's founders launched an internal-tools builder for Python developers on 2023-12-05 (Show HN, 207 points): a self-hosted middle ground between Airplane and Retool with declarative UI and Python scripts. Seven months later the 2024-07-12 Show HN introduced Dropbase AI, a significant rework: the team redesigned the product so any web app is defined by two files — properties.json for the UI and main.py for backend logic — which made it tractable for LLMs to write and regenerate reliably.
What happened
In the launch thread the founders explained the engineering choices behind the bet: regenerate main.py and show changes in a diff viewer so developers can catch hallucinations; start from boilerplate and feed the model rules about the framework to keep output bounded; and avoid generating HTML, JS or React from scratch, mapping UI behavior through properties.json instead. They said consistent results required GPT-4o or Claude 3.5 Sonnet, compared the experience to Claude Artifacts but for fully functional apps with databases and external API calls, and demoed Salesforce-leads and HubSpot-contacts editors. Commenters asked whether it could generate Django or Litestar apps; the answer was no — apps run on Dropbase's framework, though existing Python code and any PyPI package can be reused.
How it ended up
Still live as open source as of the 2026-09-05 write-up: the 2026-09-04 crawl of DropbaseHQ/dropbase shows 1.3k stars, 47 forks and 84 commits, and the README still leads with the AI-based, local-first, self-hosted builder, linking the website, docs and Discord. The record contains no shutdown, acquisition, funding or revenue data.
Background
Dropbase is an AI-based Python web app builder aimed at the internal tools developers build every day — admin panels, database editors, back-office tools and billing dashboards. A developer describes the app in natural language and the model writes Python code that runs on Dropbase's own web framework with prebuilt UI components; every change is shown in a diff viewer, can be edited by hand, and previews immediately.
The bet is that low-code and no-code builders lose because they confine app logic to UI forms and carry steep learning curves, while a tool that generates real, readable Python — local-first, self-hosted, with no credentials shared with the vendor — keeps developers in control. The founders were explicit: no black-box apps and no promised 'AI software engineer', just simple Python code developers can verify and correct.
The arc began on 2023-12-05 with a Show HN for the pre-AI internal-tools builder (207 points, 114 comments), then jumped to the 2024-07-12 Show HN for Dropbase AI (141 points, 11 comments) after the team rearchitected every app into two files — properties.json for UI and main.py for logic — to make LLM output reliable. The founders described engineering hacks (boilerplate starts, framework rules fed to the model, no from-scratch frontend generation) and said GPT-4o and Claude 3.5 Sonnet were the first models consistent enough to ship on.
The product stayed live as an open-source, self-hosted builder: the 2026-09-04 crawl of DropbaseHQ/dropbase recorded 1.3k stars, 47 forks and 84 commits, with the README still presenting the AI builder and linking the hosted website, docs and Discord. The record carries no funding, revenue, shutdown or acquisition data.
What has to be true
- Verifiable attention: three HN launches by the same founders (2023-12-05: 207 points, 114 comments; 2024-07-12: 141 points), plus a repo at 1.3k stars by the 2026-09-04 crawl.
- The bet is sharp and testable: AI-generated code that developers can read and edit versus black-box generation, with the two-file architecture as the mechanism that made LLM output reliable.
- Open source plus self-hosting answers the internal-tools trust objection directly: data and credentials stay on the developer's machine, and app folders are portable rather than locked in.
- The founders chose a bounded wedge — internal tools with table-plus-actions layouts — and said so publicly, instead of claiming to build anything.
What can be applied
When AI builds software for developers, ship the code, not the black box: editable diffs turn model errors into a review step, and open, self-hosted code makes lock-in the competitor's problem.
Aftermath
As of 2026-09-05 DropbaseHQ/dropbase remains live as an open-source project: the 2026-09-04 crawl shows 1.3k stars, 47 forks and 84 commits, and the README still describes the AI-based, local-first, self-hosted builder with demos plus links to dropbase.io, docs and Discord. The record contains no funding, revenue, shutdown or acquisition disclosures, so whether the bet became a business rather than an open-source project is not verifiable from these materials.
Sources
- DropbaseHQ/dropbase — AI-Based Python Web App Builder (1.3k stars at 2026-09-04 crawl)
- Show HN: Dropbase AI – A Prompt-Based Python Web App Builder
- Show HN: Dropbase – Build internal web apps with just Python
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card