An open source AI app builder is a tool you can self-host, read the code of, and run on your own infrastructure, where you describe an app in plain language and an AI agent builds a real, working full-stack application for you. In 2026 this category matured fast. You no longer have to choose between a closed prototype tool that traps your code on a vendor's servers and writing everything by hand. This guide compares the eight best open source AI app builders in 2026, explains what "open source" actually buys you, and shows developers exactly how to run a self-hosted, prompt-to-production builder in a few commands.
If you want a hands-on path, our tutorial on how to build an AI app builder with Next.js and one API key walks through cloning the open-source starter and deploying it.
We verified every capability claim below against each project's live repository or documentation, and re-verified every license, star count and last-commit date against the GitHub API on September 13, 2026. Where a project is young, narrow, or no longer actively maintained, we say so. Two of the eight have not shipped a commit in months, and we name them rather than quietly leaving them in the list.
If you are moving off Replit rather than self-hosting, see our roundup of the best Replit alternatives in 2026.
Quick Answer
The best open source AI app builder depends on what you are optimizing for:
- Run it locally on your desktop, bring your own model: Dyad.
- Self-host a full prompt-to-production builder you can rebrand and embed: ai-app-builder-open (the open reference frontend for the Totalum VCaaS API).
- A browser-based open version of Bolt: bolt.diy.
- Rebuild any website as editable React: Open Lovable.
- Open-source low-code for internal tools with AI: Appsmith and ToolJet.
- Build AI apps in pure Python: Reflex.
- A minimal open agent reference to learn from: app.build (see the maintenance note below before you rely on it).
- An open source no code app builder for business users: Appsmith or ToolJet, which are the two open source low code app builder platforms most teams land on.
- An open source AI website builder rather than an app builder: Open Lovable, which rebuilds an existing site as React.
If you want the shortest path from "clone a repo" to "a deployed app with auth, a database, hosting, and a custom domain," start with ai-app-builder-open, get a free key, and run it. There is a hands-on walkthrough near the end of this article.
Best open source AI app builder by use case
Most people arrive here from one of two questions: how do I set up a local-first builder on my own computer, or which open source tool should I use to build internal business tools. Here is the short answer for each, plus the cases in between.
- Best open source AI app builder for a local-first, private setup: Dyad. Runs on your machine, brings your own model keys, nothing leaves your laptop.
- Best open source AI app builder for internal business tools: Appsmith if you want the most widely deployed option, ToolJet if you prefer its workflow model. Check the license section below first, because they are not licensed the same way.
- Best open source AI app builder for a deployed, production app: ai-app-builder-open, the open reference frontend for the Totalum API. It is the only project in this guide that ships auth, a managed database, hosting, deployment and custom domains as part of the build rather than as separate integrations you wire yourself.
- Best open source AI app builder to embed and rebrand inside your own SaaS: ai-app-builder-open, because it is MIT licensed and multitenant. Most of the alternatives are either single-user by design or carry license terms that specifically restrict multi-tenant use.
- Best open source AI app builder for Python teams: Reflex.
- Best open source AI app builder for agentic workflows rather than apps: Giselle or Dify.
- Best open source AI website builder: Open Lovable, with the maintenance caveat noted below.
What is an open source AI app builder?
An open source AI app builder is software, released under a license like MIT, that turns a natural-language prompt into a working software application, and whose source code you are free to read, modify, self-host, and in many cases resell. Three plain-English terms matter here:
- Self-hosting means you run the software on your own servers or laptop instead of only using it as a website someone else controls.
- Full-stack means the generated app includes both the visible interface and the behind-the-scenes parts: a database, user login, and server logic.
- Prompt-to-production means you go from a sentence describing your idea to a live URL your users can visit, not just a preview.
The reason this category exists is ownership. Closed tools like Lovable, Bolt, v0, and Replit are excellent at fast prototypes, but the output usually lives inside their platform. An open source builder gives you the code, the ability to host it anywhere, and, for the API-driven ones, the ability to put the whole builder inside your own product.
Who this is for
If you are a non-technical founder, the takeaway is simple: open source builders let you avoid lock-in and keep the app you paid to create. You can hand the code to any developer later. If you are a developer or an agency, open source builders are the difference between renting a feature and owning a platform you can white-label. White-labeling means putting your own brand on a tool built on someone else's engine.
How we picked
We looked for projects that are genuinely open source (a real license, public code), actively maintained in 2026, and aimed at building applications rather than just chatbots or static sites. We weighted four things: whether you can self-host, whether it produces a real full-stack app, whether it is production-capable rather than demo-only, and whether you can embed or rebrand it. We note honest tradeoffs for each.
The 8 best open source AI app builders in 2026
1. Dyad
Dyad is the most visible name in the "open source AI app builder" search results, and for good reason. It is a local-first desktop app: it runs on your own computer, you bring your own AI model keys, and you own the code it generates with zero lock-in. It is Apache 2.0 licensed, not MIT, and the license carries a carve-out worth knowing about: everything outside the repository's src/pro/ directory is Apache 2.0, while content inside src/pro/ is governed by a separate license defined in that directory. In practice Dyad is an open core project rather than a uniformly permissive one. Verified against the repository's LICENSE file on September 13, 2026 (21,508 stars, last commit September 11, 2026, actively maintained).
Best for: developers who want a private, local builder and already have model API keys.
Honest tradeoff: because it is desktop and local-first, it is not designed to be a multitenant, hosted builder you embed for your own customers. It is a tool for you, not a platform you resell.
2. ai-app-builder-open (the open Totalum builder)
ai-app-builder-open is an MIT-licensed, self-hostable prompt-to-full-stack-app builder built on Next.js 16, React 19, TypeScript 5, and Tailwind CSS 4. From a single prompt it generates a real application and gives you a chat panel, live preview, an integrated Monaco code editor, a managed database with a browsing UI, secrets management, one-click hosting, custom domains, bidirectional GitHub sync, version history with rollback, live server logs, isolated sandboxes, a visual git-diff viewer, and multitenant architecture out of the box.
The heavy lifting (the agent, sandboxes, hosting, database, and deployment) runs through the Totalum VCaaS API, which the app reaches with one environment variable, TOTALUM_VCAAS_API_KEY. That design is what makes it interesting for two audiences at once: an individual can self-host a complete builder, and a SaaS company or agency can use the same engine to embed a multitenant, white-label builder inside their own product.
Best for: anyone who wants a self-hosted, embeddable, prompt-to-production builder they can rebrand, and teams evaluating an AI app builder they can embed via API.
Honest tradeoff: it is a young project, so its community is smaller than Dyad's today, and it depends on the Totalum API for the build engine rather than running fully offline. That dependency is the point (it is what delivers hosting, a managed database, and deployment), but it is a real design choice to weigh.
3. bolt.diy
bolt.diy is the community, open-source sibling of Bolt.new. It runs in the browser, supports multiple model providers, and lets you generate and iterate on full-stack apps in a web IDE. It is a strong pick if you like the Bolt workflow but want an open version.
It is MIT licensed and carries 19,867 stars, verified September 13, 2026.
Best for: people who want the Bolt-style, in-browser experience without the closed platform. See our take on Bolt.new alternatives in 2026.
Honest tradeoff: the browser sandbox model can be limiting for larger or more custom backends, and the project has not received a commit since February 7, 2026. Seven months of silence on a fast-moving category is a real risk, not a detail. The stars are historical; check the commit history yourself before you build on it.
4. Open Lovable
Open Lovable, from the Firecrawl team, is an open-source app that takes a website URL and rebuilds it as a clean, editable React app. It is narrower than a general builder, but excellent for that specific job.
It is MIT licensed with 28,400 stars. Note that the repository moved organisations, from mendableai/open-lovable to firecrawl/open-lovable, so older links and older listicles point at a redirect. Verified September 13, 2026.
Best for: quickly turning an existing site or design into React code you own. This is also the closest thing in this guide to an open source AI website builder, as distinct from an app builder.
Honest tradeoff: it is focused on the rebuild-from-URL use case, not on being a full hosting-and-database platform, and its last commit was November 19, 2025, which makes it the least actively maintained project we still recommend. If you are comparing it to the closed original, see Lovable alternatives in 2026.
5. Appsmith
Appsmith is a mature, widely adopted open-source low-code platform for building internal tools and admin panels, with AI features layered on top. It connects to your databases and APIs and is battle-tested in production at many companies.
It is Apache 2.0 licensed (40,865 stars, last commit September 11, 2026). Apache 2.0 is the permissive end of the spectrum, which matters if you intend to build a commercial product on top of it.
Best for: internal business tools, dashboards, and admin panels. If you searched for an open source no code app builder or an open source low code app builder, this is usually the tool you meant.
Honest tradeoff: it is a low-code drag-and-connect platform first and an AI generator second, so it is a different shape from the prompt-to-app tools above.
6. ToolJet
ToolJet is another established open-source low-code option for internal apps and workflows, with a visual builder and AI-assisted features. Like Appsmith, it is strong for data-connected internal software.
It is AGPL-3.0 licensed (40,897 stars, last commit September 12, 2026). That is the single most important difference between ToolJet and Appsmith, and almost no comparison article mentions it. AGPL is a strong copyleft license: if you modify ToolJet and offer it to users over a network, the license requires you to make your modified source available to those users. For internal use that is usually irrelevant. For a commercial product built on a fork, it is a decision you want your lawyer in the room for.
Best for: teams standardizing internal tooling on an open platform.
Honest tradeoff: same category caveat as Appsmith. It is internal-tool focused rather than a customer-facing app generator, and the AGPL terms make it the hardest project here to build a closed commercial product on.
7. Reflex
Reflex is an open-source framework for building full web apps in pure Python, with AI-assisted generation. If your team lives in Python, it removes the JavaScript context switch entirely.
It is Apache 2.0 licensed (28,881 stars, last commit September 13, 2026, the most recently active project in this guide).
Best for: Python teams and data or ML engineers who want to ship web apps without a separate frontend stack.
Honest tradeoff: it is a framework you code in, so it expects more engineering than a one-prompt builder.
8. app.build
app.build is a minimal, open-source reference implementation of an AI app-building agent. It is less a product and more a clean codebase to study and extend if you want to understand how these systems work under the hood.
Best for: engineers who want to learn the internals or fork a starting point.
Honest tradeoff, and it is now a serious one: we could not verify an actively maintained public repository for app.build as of September 13, 2026. The appdotbuild organisation's only public repository carries 6 stars, no license detected by GitHub's license API, and a last push of March 19, 2026, and the app.build homepage returns a 2.5 KB shell with no readable content. We are leaving the entry in place because the original codebase was genuinely instructive, but treat it as dormant and do not start a new project on it. If you know of an active mirror, we will update this entry.
Comparison table
Every license identifier, star count and last-commit date below was read from the GitHub API on September 13, 2026. We publish the last-commit column because "open source" tells you nothing about whether anyone is still maintaining the project, and in this category that turns out to matter a great deal.
| Builder | License (verified) | Stars | Last commit | Self-host | Multitenant / embeddable | Best for |
|---|---|---|---|---|---|---|
| Dyad | Apache-2.0, with a proprietary src/pro/ carve-out |
21,508 | Sep 11, 2026 | Yes (desktop, local-first) | No | Private local builds, bring your own model |
| ai-app-builder-open | MIT | 54 | Sep 11, 2026 | Yes | Yes | Self-hosted, rebrandable prompt-to-production builder |
| bolt.diy | MIT | 19,867 | Feb 7, 2026 | Browser-based | No | Open Bolt-style workflow |
| Open Lovable | MIT | 28,400 | Nov 19, 2025 | Yes | No | Rebuild any site as React |
| Appsmith | Apache-2.0 | 40,865 | Sep 11, 2026 | Yes | Partial | Internal tools and admin panels |
| ToolJet | AGPL-3.0 | 40,897 | Sep 12, 2026 | Yes | Partial | Internal apps and workflows |
| Reflex | Apache-2.0 | 28,881 | Sep 13, 2026 | Yes | N/A (framework) | Full web apps in Python |
| app.build | none detected | 6 | Mar 19, 2026 | Yes | No | Dormant, see the note above |
Three things in that table are worth stopping on, because they contradict what most comparison articles in this category still say. Dyad is Apache 2.0, not MIT, and it is an open core project with a proprietary directory. ToolJet is AGPL, not a permissive license, which is the single biggest practical difference between it and Appsmith. And two of the eight have been silent for months, which no star count will tell you.
Every project in this table is genuinely open. The clearest split is between builders that generate an app for you to own (Dyad, ai-app-builder-open, bolt.diy, Open Lovable) and low-code platforms you assemble apps in (Appsmith, ToolJet), with Reflex sitting apart as a code-first framework.
For developers: run a self-hosted AI app builder in 4 commands
Here is the fastest path from nothing to a running, self-hosted builder. This uses ai-app-builder-open because it ships as a complete Next.js app.
First, clone and install:
git clone https://github.com/totalumlabs/ai-app-builder-open.git
cd ai-app-builder-open
npm installNext, get an API key. Create a Totalum account, choose "Use the Totalum API," and copy your key. The key is what unlocks the build engine: the AI agent, sandboxes, the managed database, hosting, and deployment. Keep this key server-side only; the app already proxies it through its backend so it never reaches the browser. The Totalum quickstart and the API overview walk through getting a key, and the build-your-own-AI-app-builder guide explains the reference frontend in detail.
Then set the environment variable and start the dev server:
export TOTALUM_VCAAS_API_KEY="your_key_here"
npm run devOpen the local URL, type a prompt like "a booking app for a barber shop with login and a calendar," and watch the agent scaffold a full-stack app. From the panels you can browse the generated code in Monaco, inspect and edit the database, view live logs, roll back versions, connect a GitHub repo, and attach a custom domain when you deploy. For non-developers reading along: this is the moment the "describe it, get a real app" promise becomes concrete, and because it is self-hosted and MIT licensed, the result is genuinely yours.
Open source no code app builders and low code app builders
If you searched for an open source no code app builder or an open source low code app builder, you are probably not looking for the same thing as someone searching for an AI app builder, and the distinction is worth making explicitly because the two categories get blended constantly.
- A no code app builder is one you operate entirely through a visual interface. You drag components, connect a data source, and configure behaviour in forms. No text file is ever edited.
- A low code app builder is mostly visual but expects you to drop into small snippets of code (usually JavaScript or SQL) for the parts the interface does not cover.
- An AI app builder generates the application from a written prompt. The output is real code, and how much of it you can see, edit and take with you varies enormously by tool.
In the open source world these categories map cleanly onto specific projects:
| What you searched for | The open source projects that actually match |
|---|---|
| Open source no code app builder | Appsmith, ToolJet |
| Open source low code app builder | Appsmith, ToolJet, and Reflex if you count Python as low code |
| Open source AI app builder | Dyad, ai-app-builder-open, bolt.diy, Open Lovable |
| Open source AI website builder | Open Lovable (rebuilds an existing site as React) |
| Self-hosted AI app builder with hosting and a database | ai-app-builder-open |
The honest summary: there is no mature open source no code app builder that also generates apps from a prompt and hosts them for you. Appsmith and ToolJet are excellent visual builders with AI features layered on, and the prompt-to-app projects are code generators. ai-app-builder-open is the closest to both at once, because the prompt produces a real deployed application, but you are still handling a repository rather than working only in a visual canvas.
Three more open source projects you will see named
These are not in our top eight, because they are either not app builders in the sense most readers mean or they are aimed at a different job. They come up constantly in this category, so here is where each one actually fits. All figures verified September 13, 2026.
Giselle (github.com/giselles-ai/giselle, Apache-2.0, 555 stars, last commit September 1, 2026) is an open source platform for agentic workflows rather than for generating applications. You self-host it and connect it to the major model providers. If what you want is a chain of AI steps collaborating with humans, Giselle is the right shape. If what you want is a deployed app with a login screen, it is not.
Dify (github.com/langgenius/dify, 155,560 stars, last commit September 12, 2026) is by far the largest project mentioned here, and it is an LLM application development platform: agents, workflows, RAG pipelines and model orchestration. Read its license carefully before you build a business on it. Dify is not plain Apache 2.0. It uses a modified Apache 2.0 with additional conditions, and the most consequential one is explicit: unless Dify authorises it in writing, you may not use the source to operate a multi-tenant environment. There is also a clause preventing you from removing Dify's logo and copyright information from the frontend. If your plan was to self-host Dify and resell it to your own customers, that plan requires a commercial license.
OpenUI (github.com/wandb/openui, Apache-2.0, 22,543 stars, last commit September 10, 2026) from Weights and Biases generates UI components from a description and lets you convert them to React or other frameworks. It is a genuinely useful open tool for the interface layer, and it is not trying to be a full-stack builder.
License check: which of these can you actually embed and resell?
This is the question that separates a weekend project from a business, and it is the one almost no comparison article answers. "Open source" is not one thing. Here is what each license in this guide actually permits if your intention is to put the builder inside your own product and charge for it.
| License | Projects | Can you build a closed commercial product on it? |
|---|---|---|
| MIT | ai-app-builder-open, bolt.diy, Open Lovable | Yes. Keep the copyright notice and you are free to modify, rebrand and sell. |
| Apache-2.0 | Appsmith, Reflex, Giselle, OpenUI | Yes, with an explicit patent grant, which is a point in its favour. Note attribution and change-notice requirements. |
| Apache-2.0 with a proprietary carve-out | Dyad | Mostly, but the src/pro/ directory is governed separately. Read that directory's license before shipping anything derived from it. |
| AGPL-3.0 | ToolJet | Only with care. If you modify it and offer it over a network, you must publish your modified source to those users. |
| Modified Apache-2.0 with conditions | Dify | Not for multi-tenant use without a written commercial license from Dify. Logo and copyright removal is also prohibited. |
| None detected | app.build | No. Code with no license is, by default, all rights reserved. |
For the specific case of embedding a prompt-to-app builder inside a SaaS product and charging your own customers for it, the permissive MIT projects are the only ones in this guide that are unambiguously clear, and of those only ai-app-builder-open is multitenant and ships a production backend. That is not a coincidence, it is what it was built for. If you are evaluating this seriously, our guide on how to embed an AI app builder in your SaaS via API covers the architecture, and white-label AI app builders covers the commercial side.
This is not legal advice. Licenses change, and the only authoritative source is the LICENSE file in the repository on the day you read it. We re-checked every one of them on September 13, 2026.
Is there a completely free open source AI app builder?
Yes, and more than one, but "free" splits into two different things that are easy to conflate.
Free to run, with no vendor bill at all: Dyad is the clearest example. You install it, you supply your own model API keys, and Dyad itself never charges you. Your only cost is what OpenAI, Anthropic or your local model setup charges for tokens. bolt.diy works the same way. If you run a local model through Ollama or LM Studio, the marginal cost genuinely is zero.
Free and open source software, with a paid service behind part of it: ai-app-builder-open is MIT licensed and free to clone, fork and rebrand forever, but the build engine it calls (the agent, sandboxes, managed database, hosting and deployment) runs through the Totalum API, which has a free tier and then paid plans. You are not paying for the software. You are paying for the infrastructure that turns a prompt into a running, hosted application, which is the part that cannot be free because it costs real money to run.
The tradeoff is straightforward. Fully free tools give you generated code on your own machine and leave hosting, the database, auth and deployment to you. Tools with a service behind them hand you a live URL. Neither is better in the abstract; it depends on whether the work you want to skip is the infrastructure work.
How to choose
If self-hosting is a preference rather than a hard requirement, it is worth weighing these against the hosted field too: our AI app builder comparison for 2026 ranks every major option on starting price, backend, and whether you own the code.
- You want a private tool just for you: Dyad.
- You want to ship a real app with hosting, a database, and a domain, fast: ai-app-builder-open, and consider the wider field in our production-ready AI app builders comparison.
- You want to embed a builder inside your own SaaS and rebrand it: ai-app-builder-open on the Totalum API, and read the best AI app builders for SaaS in 2026 and white-label AI app builders.
- You want internal tools: Appsmith or ToolJet.
- You live in Python: Reflex.
If you are weighing open source against the well-known closed tools, our comparisons of Lovable vs Bolt vs Totalum and free AI app builders in 2026 are useful next reads.
For a closer head-to-head on one of these local, open-source tools, see our Dyad AI app builder vs Totalum comparison.
If your goal is to embed one of these engines inside your own SaaS, see how to use an AI app builder API to add a prompt-to-app builder to your product.
If you would rather stay on a hosted platform but still keep a copy of the source, compare the hosted options first: we break down every AI app builder that lets you export your code, including which ones put the download behind a paid tier.
FAQ
Which AI app builder can build an app for free?
Several here are free and open source, including Dyad and bolt.diy, which you run yourself. ai-app-builder-open is free and MIT licensed to self-host; the Totalum API it calls includes free credits to start, then moves to paid usage once you scale. There is no cost to clone the repo and try it.
Can I build my own AI app builder?
Yes. That is exactly what ai-app-builder-open is designed for. It is an open reference frontend for the Totalum VCaaS API, so you can fork it, put your own brand on it, and run it as your own product. The build-your-own-AI-app-builder guide covers the setup.
What is the difference between an open source AI app builder and Lovable, Bolt, or v0?
The closed tools are polished and fast for prototypes, but the code and the app typically stay on their platform. An open source AI app builder gives you the source, the freedom to self-host anywhere, and, for the API-driven ones, the ability to embed and resell the builder itself.
Can I embed an open source AI app builder inside my own SaaS?
Yes, if it is API-driven and multitenant. ai-app-builder-open plus the Totalum VCaaS API is built for this: your customers generate and deploy apps without leaving your product, under your brand. See the API overview and our guide on how to embed an AI app builder via API.
What is the best free AI app builder?
For a genuinely free, no-vendor-bill setup, Dyad is the best answer in 2026: local-first, Apache 2.0, and you supply your own model keys. If "free" means free to start and you want hosting, a database and a deployed URL included, ai-app-builder-open is free and MIT licensed to self-host and the Totalum API behind it includes free credits before any paid plan. Our full comparison of free AI app builders in 2026 covers the hosted options too.
Can I build an app with AI for free?
Yes. Clone Dyad or bolt.diy, point them at a model you already pay for or a local model through Ollama, and you can generate a working application without paying either project anything. The cost shows up later, at hosting and database time, which is the step the free tools leave to you.
Is there an open source AI app builder on GitHub I can just clone?
All of them are on GitHub. The ones actively maintained as of September 13, 2026 are Dyad, ai-app-builder-open, Appsmith, ToolJet, Reflex, Giselle, Dify and OpenUI. bolt.diy (February 2026) and Open Lovable (November 2025) have not received commits in months, and app.build appears dormant.
Is there an open source AI mobile app builder?
Not a mature one. Every project in this guide targets web applications. The realistic 2026 route to a mobile app from an open source builder is to generate a responsive web application and wrap or install it as a progressive web app, rather than to expect native iOS and Android output.
Is ai-app-builder-open really production-ready?
The generated apps include hosting, a managed database, auth, custom domains, and deployment, which are the pieces a prototype tool usually leaves out. It is a young project, so evaluate it against your needs, but it is built for real deployment rather than demos.
The bottom line
Open source AI app builders in 2026 fall into three groups: private local tools (Dyad), owned-code app generators (ai-app-builder-open, bolt.diy, Open Lovable), and open low-code platforms (Appsmith, ToolJet), with Reflex as a Python-first framework. If you want the shortest route to a deployed, ownable app, and the option to embed the whole builder in your own product later, ai-app-builder-open is the one to clone first.
Correction, September 13, 2026: an earlier version of this article stated that Dyad is MIT licensed. Dyad is licensed under Apache 2.0, with content in its src/pro/ directory governed by a separate license defined in that directory. The comparison table has been corrected and every other license claim in this guide was re-verified against the GitHub API on the same date.
If this is useful, star the repository at github.com/totalumlabs/ai-app-builder-open so more builders can find it, and create a free Totalum account to get an API key and build your first app in minutes. Building a SaaS and want the engine inside your own product? Start with the Totalum API overview.