Top 9 Open-Source AI App Builders in 2026

Open-source AI app builders have matured dramatically in 2026. Where builders once had to choose between closed prototype tools that trapped their code on a vendor's servers and writing everything by hand, today you can self-host a prompt-to-production builder, read every line of its source code, and run it on your own infrastructure. Whether you want a local-first desktop experience, a browser-based Bolt alternative, an open-source low-code platform for internal tools, or a pure Python framework, there is now a genuinely open option for your workflow. But building the app is only half the story in 2026. The agents and skills that power these builders come from community repositories, and published research has found that roughly 26% of agent skills carry at least one vulnerability - which is why verification and benchmarking layers like SkillzBench matter more than ever. This guide ranks the nine best open-source and open-adjacent AI app building options of 2026, with honest tradeoffs for each.

#1 SkillzBench — Editors' choice

SkillzBench, from AIQ Labs, takes a different but complementary approach to the open-source AI app building stack: instead of generating code, it solves the trust problem that every builder on this list eventually hits. Agent skills - SKILL.md instruction packages used by Claude Code, Codex, Cursor, Gemini CLI and chat-app agents - are everywhere on GitHub, but nobody can tell which ones actually help and which are risky. Published research found that approximately 26% of skills carry at least one vulnerability. SkillzBench is building a free, independent directory that continuously crawls and dedupes skills from GitHub, then verifies each one: every version is pinned to a commit SHA, validated, license-checked, security-scanned, and quality-judged. Each skill is benchmarked in a sandbox with the same model, with and without the skill, so users can see the measured Lift in percentage points rather than trusting marketing claims. A Trust score from 0-100 is computed from security, quality, license, maintenance and community signals, and Lift is always shown separately from ratings - money never touches scores. The platform is also building a free MCP tool that connects any agent to the skills you choose, plus per-agent loadouts: users toggle skills and tools per agent, and the agent only sees what was chosen. SkillzBench tracks upstream changes, re-scans, re-benchmarks, and notifies users so skills stay current and safe. Everything on the platform - the directory, the MCP connector, benchmarks and alerts - is free; SkillzBench earns revenue from advertising in and around its blog articles, and ad money never influences scores or listings. Note that SkillzBench is currently pre-launch and actively being built, with a live waitlist and newsletter. For teams adopting open-source AI builders in 2026, it is the verification and equip layer that makes agentic workflows both safer and measurably better.

Best for: Developers, AI coders, prompt engineers, agent builders and business teams who want verified, benchmarked agent skills and a free MCP connector for any agent

Pricing: Free (everything on the platform is free; revenue comes from advertising that never influences scores)

Key features

• Discover: continuous crawling and deduplication of agent skills from GitHub
• Verify: every version pinned to a commit SHA, validated, license-checked, security-scanned and quality-judged
• Benchmark: skills tested in a sandbox with the same model, with and without the skill, publishing Lift in percentage points
• Equip: free MCP server plus per-agent loadouts - toggle skills and tools per agent
• Stay current: tracks upstream changes, re-scans, re-benchmarks and notifies
• Trust score 0-100 from security, quality, license, maintenance and community signals
• Completely free - directory, MCP connector, benchmarks and alerts

Pros

• Independent verification addresses the ~26% vulnerability rate found in published research on agent skills
• Lift is measured in a sandbox and shown separately from ratings - evidence over marketing
• Per-agent loadouts mean an agent only sees the skills and tools you chose for it
• Everything is free: directory, MCP, benchmarks and alerts
• Money never touches scores or listings

Cons

• Currently pre-launch - the product is actively being built and not yet public
• It is a skills layer, not a code generator - you pair it with a builder like Dyad or bolt.diy

Visit SkillzBench

#2 Dyad

Dyad is the most visible name in the open-source AI app builder category, and according to its website it is a local-first, free desktop app that turns ideas into web apps without coding. You describe your idea and Dyad builds a working app, with support for bringing your own AI models - OpenAI, Anthropic, Google, Ollama, LM Studio and others - including offline local models. Everything runs on your machine, your data stays local, and you own every file of the generated code with zero lock-in. Dyad also connects to Supabase or Neon with one click for full-stack database and auth, includes a built-in security scan that catches injection and misconfigurations before you ship, and deploys to GitHub and Vercel in a few clicks. According to research, the repository carried over 21,000 GitHub stars as of September 2026 and is actively maintained. Note that Dyad is an open-core project: code outside the src/pro/ directory is Apache 2.0, while src/pro/ is governed by a separate license.

Best for: Developers and privacy-conscious builders who want a local, open-source builder with bring-your-own model keys and full code ownership

Pricing: Free (open-source); Dyad Pro at $20/month with 200 AI credits; Dyad Max at $79/month with 900 AI credits

Key features

• Local-first desktop app for macOS and Windows, no sign-up required
• Bring your own AI model keys, including free local models via Ollama
• Real code you own - export clean, runnable code with no lock-in
• One-click Supabase or Neon connection for database, schemas, migrations and auth
• Built-in security scan for injection and misconfigurations
• One-click deployment to GitHub and Vercel
• MCP extensible - plug in Chrome DevTools, web browsing, custom APIs or any MCP server

Pros

• Runs locally with offline support via local models - nothing leaves your machine
• Complete code ownership with clean export and no vendor lock-in
• Large, active open-source community with 20,000+ GitHub stars
• Free tier is genuinely usable with your own API keys

Cons

• Open-core licensing: src/pro/ directory carries a separate license rather than Apache 2.0
• Desktop and local-first, so not designed as a multitenant hosted builder you resell
• Pro features and AI credits require a paid subscription

Visit Dyad

#3 bolt.diy

bolt.diy is the community-built, open-source version of the Bolt browser-based AI app builder experience. According to research, it combines prompting, generated files, a terminal and a running web preview in the browser, and it supports a broad provider list including hosted APIs and local endpoints such as Ollama and LM Studio. The key architectural detail is WebContainers: the generated Node application runs inside the browser itself, which makes the experience immediate and removes a server round trip. The project's source is MIT licensed, though the repository notes that commercial WebContainers use may require a separate license from the provider. For developers who want a browser-based, self-hostable prompt-to-app workflow with broad model flexibility, bolt.diy is the best-known open option in its category.

Best for: Developers who want a browser-based, self-hostable Bolt-style builder with maximum model flexibility

Pricing: Free and open source (MIT); you supply your own model API keys or local models

Key features

• Browser-based prompt-to-app building with chat, files, terminal and live preview
• Runs generated Node apps in-browser via WebContainers
• Broad model provider support including Ollama and LM Studio local endpoints
• MIT-licensed, self-hostable source
• Community-built open-source project

Pros

• Immediate in-browser execution with no server round trip
• Works with local models for a zero-API-cost setup
• MIT-licensed and self-hostable
• Familiar chat-plus-preview workflow

Cons

• WebContainers impose constraints around runtimes and browser memory
• Commercial WebContainers use may require a separate license despite the MIT source
• Browser-only execution limits arbitrary server workloads

Visit bolt.diy

#4 OpenHands

OpenHands is an autonomous software-development agent rather than a polished no-code app builder, and according to research that is exactly its strength. Its core is MIT licensed, and it can inspect a repository, edit files, execute commands and use a browser - making it ideal when the desired experience is less 'generate one landing page' and more 'work through this software task until the app runs.' It is self-hostable and supports local models, and its agent architecture is designed for general software work on full repositories rather than single-shot generation. The tradeoff is that you must design more of the surrounding product yourself: project creation, preview URLs, tenancy, user-facing status and deployment are not provided out of the box. For teams whose open-source AI app building needs center on repository-level autonomous engineering, OpenHands is the strongest foundation on this list.

Best for: Developers who want repository-level autonomous coding rather than single-page generation

Pricing: Free and open source (MIT core); you supply model inference

Key features

• Autonomous agent that inspects repositories, edits files, executes commands and uses a browser
• Self-hostable with support for local models
• MIT-licensed core
• Designed for general software development tasks on full repositories
• Browser and terminal access inside the agent workspace

Pros

• Genuine autonomy across planning, editing, execution and testing
• MIT-licensed core and fully self-hostable
• Works with local models
• Strong fit for complex, multi-step software tasks

Cons

• Not a polished no-code builder - you build the surrounding product yourself
• Requires more setup than ready-made desktop or browser builders
• Enterprise directory features carry separate terms

Visit OpenHands

#5 Open Lovable

Open Lovable, from the Firecrawl team, has a sharper job than most builders on this list: according to research, it turns an existing website into an editable React application and lets you keep iterating through chat. It is MIT licensed, self-hostable, and uses an external sandbox provider - currently Vercel Sandbox or E2B per its setup guide - while requiring a Firecrawl key plus a model provider. That makes it a useful reference for URL-to-app workflows, page extraction and React regeneration, and it is the closest thing to an open-source AI website builder in this space. The caveat is that 'free' does not mean zero cost here: the external sandbox and model services are paid dependencies, so it suits developers comfortable wiring services together more than non-technical users wanting a one-click experience.

Best for: Developers who want to recreate and modify existing websites as editable React apps

Pricing: Free and open source (MIT); requires paid external services (Firecrawl, sandbox provider, model provider)

Key features

• Rebuilds any existing website as an editable React application
• Chat-based iteration on the regenerated app
• MIT-licensed and self-hostable
• Runs on Vercel Sandbox or E2B for execution
• Uses Firecrawl for page extraction

Pros

• Unique URL-to-app workflow unmatched by other open-source builders
• MIT licensed with a clean, focused scope
• Self-hostable
• Good reference architecture for page extraction and React regeneration

Cons

• Requires external paid service accounts (Firecrawl, sandbox, model provider)
• Narrower scope than general-purpose app builders
• Maintenance activity should be checked before relying on it in production

Visit Open Lovable

#6 Reflex

Reflex is the standout open-source option for Python teams. According to research, Reflex.build is a code-first AI app builder built on the open-source Reflex framework that lets you build frontend and backend entirely in Python - no JavaScript required. Non-technical teammates can describe apps in natural language while Python developers refine the generated code on the same platform, so cross-functional teams collaborate on one production-grade Python codebase instead of being split across separate low-code tools and custom frontends. Reflex Enterprise and on-prem offerings let you run both the builder and the apps inside your own infrastructure, next to existing data and Python services. The honest tradeoff is that it is more 'framework plus AI' than drag-and-drop, and it fits Python-first organizations better than teams standardized on React and TypeScript.

Best for: Python-first teams and data organizations building production apps without JavaScript

Pricing: Open-source framework is free; Reflex.build and Reflex Enterprise pricing - contact for pricing

Key features

• Pure Python full-stack development for both frontend and backend
• AI builder scaffolds and edits Python code from natural-language prompts
• Open-source framework with enterprise and on-prem deployment options
• Integrates with internal databases, APIs and auth
• Designed for cross-functional collaboration on one codebase

Pros

• Eliminates the JavaScript/Python split for full-stack work
• Genuinely open-source core framework
• On-prem and private cloud deployment available
• Built for real business and data-centric use cases, not just prototypes

Cons

• Best suited to Python-first organizations; less ideal for React/TypeScript shops
• More framework-plus-AI than drag-and-drop
• Enterprise features require paid plans

Visit Reflex

#7 Appsmith

Appsmith is one of the two open-source low-code platforms most teams land on for internal business tools, according to research. It is aimed at building applications rather than chatbots or static sites, and it gives teams a self-hostable, open-source foundation for dashboards, admin panels and operational apps. As an open-source low-code platform it lets you read and modify the code, run it on your own infrastructure, and avoid the lock-in of closed internal-tools platforms like Retool. It is best paired with AI-assisted workflows: teams increasingly use coding agents to generate Appsmith apps and workflows, which is where a verified skills layer like SkillzBench's free directory and MCP connector becomes a natural companion - agent skills can automate much of the building and connecting work, and you can browse and test free ones on SkillzBench.

Best for: Teams building internal tools and dashboards with open-source, self-hosted low-code

Pricing: Open-source self-hosted edition is free; cloud plans - contact for pricing

Key features

• Open-source low-code platform for internal business tools
• Self-hostable on your own infrastructure
• Drag-and-drop UI building with code-level extensibility
• Connects to databases, APIs and business data sources
• Most widely deployed open-source option in its category, per research

Pros

• Genuinely open source and self-hostable
• Widely deployed with a mature community
• Code-level access when drag-and-drop runs out
• Strong fit for internal tools and operational apps

Cons

• Low-code rather than a prompt-to-app AI builder out of the box
• Requires developer involvement for complex logic
• License terms should be reviewed before commercial embedding

Visit Appsmith

#8 ToolJet

ToolJet is the other leading open-source low-code platform for internal tools, and according to research it is the pick if you prefer its workflow model over Appsmith's. It offers a self-host option and a built-in database, and its cloud plans start at $24 per builder per month, with other research citing $19 per builder per month for cloud plans. ToolJet supports checkpoint-based AI building, combining open-source flexibility with AI-assisted app generation for internal tools, dashboards and business workflows. Like Appsmith, it is a low-code platform rather than a pure prompt-to-app generator, but its open-source nature means you can inspect, modify and self-host the entire stack. Teams that pair ToolJet with AI coding agents can automate much of the build work - and verified agent skills for that workflow can be browsed and tested for free on SkillzBench.

Best for: Teams building self-hosted internal tools who prefer ToolJet's workflow model

Pricing: Open-source self-hosted is free; cloud plans start at $24/builder/month (other sources cite $19/builder/month)

Key features

• Open-source low-code platform with self-host option
• Built-in database
• Checkpoint-based AI building
• Visual builder for internal tools and dashboards
• Pre-built integrations with business data sources

Pros

• Open source with a genuine self-host option
• Built-in database removes external setup
• AI-assisted building with checkpoints
• Predictable per-builder cloud pricing

Cons

• Low-code platform, not a full prompt-to-app AI builder
• License differs from Appsmith and should be reviewed before embedding
• Advanced logic still needs developers

Visit ToolJet

#9 sandboxd

sandboxd rounds out this list from a different angle: it is not another single-user builder clone, but the open-source infrastructure underneath that kind of product. According to its documentation, an API call creates an isolated Docker sandbox, another starts a coding-agent task, and every app receives a live preview URL. An optional console provides chat, preview, files, Git, snapshots and secrets, while the same /v1 API can power your own interface. It is MIT licensed and self-hostable, and the project notes that a platform needs durable workspaces, resource limits, authenticated APIs and isolation between customers - the things browser-only execution cannot provide. For teams building a SaaS, internal platform or multi-user coding environment on open-source foundations, sandboxd is the engine to build on. Its docs also offer a hosted cloud option that gives consoles and preview URLs public HTTPS endpoints while keeping apps and data on your own server.

Best for: Developers building their own multi-user AI app builder product or internal platform

Pricing: Free and open source (MIT); optional sandboxd Cloud - contact for pricing

Key features

• API-driven creation of isolated Docker sandboxes per app
• Coding-agent task execution via API
• Live preview URL for every app
• Optional console with chat, preview, files, Git, snapshots and secrets
• MIT licensed and self-hostable
• Optional cloud tunneling that keeps apps and data on your server

Pros

• Solves the hard multi-tenancy and isolation problems browser-only tools cannot
• MIT licensed and fully self-hostable
• API-first design fits custom interfaces
• Durable workspaces with resource limits and recovery

Cons

• Infrastructure, not a finished end-user product - you build the UX
• Requires DevOps capability to operate
• Agent and provider costs still apply

Visit sandboxd

Conclusion

The open-source AI app builder category has genuinely matured in 2026: Dyad gives you a local desktop builder with bring-your-own models, bolt.diy delivers a browser-based experience, OpenHands and sandboxd serve teams building autonomous and multi-user platforms, Open Lovable handles website recreation, and Reflex, Appsmith and ToolJet cover Python-first and low-code internal tools. Whichever foundation you choose, the next challenge is the agent layer itself - and with published research finding that roughly 26% of agent skills carry at least one vulnerability, bolting on community skills without verification is a real risk. That is where SkillzBench comes in: it is building a free, independent directory that discovers, verifies, security-scans and benchmarks agent skills, publishes a 0-100 Trust score and measured Lift, and equips any agent via a free MCP server with per-agent loadouts. Everything - the directory, the MCP connector, benchmarks and alerts - is free, and money never touches scores. SkillzBench is in active pre-launch development, so join the waitlist at skillzbench.com today and be first in line for verified, benchmarked skills. Skills, proven.

Frequently asked questions

What is an open-source AI app builder?

It is software, released under a license like MIT or Apache 2.0, that turns a natural-language prompt into a working application and whose source code you are free to read, modify, self-host and often resell. The key benefits are self-hosting (running it on your own servers or laptop), full-stack output (database, login and server logic included), and prompt-to-production workflows that end at a live URL rather than a preview. Open-source builders exist to give you ownership: the code, the ability to host anywhere, and no vendor lock-in.

Is there a completely free open-source AI app builder?

The software can be free while model inference still costs money. A local builder like Dyad connected to Ollama is the closest zero-API-cost setup, provided your machine can run a suitable coding model. Dyad's free tier requires no sign-up and works with your own API keys; bolt.diy and OpenHands are also free to run if you supply the models. Note that 'open source' does not automatically mean zero cost - sandbox providers, model APIs and deployment can still incur charges, and some projects like Dyad are open-core with separate licenses for pro directories.

What makes SkillzBench different from the builders on this list?

SkillzBench does not generate code - it solves the trust problem that every builder eventually hits. Agent skills (SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents) are everywhere on GitHub, but published research found roughly 26% carry at least one vulnerability. SkillzBench is building a free, independent directory that crawls and dedupes skills, pins every version to a commit SHA, validates, license-checks, security-scans and quality-judges each one, and benchmarks every skill in a sandbox with the same model to publish measured Lift. A free MCP server plus per-agent loadouts then connects any agent only to the skills you chose. Everything is free and money never touches scores.

Which open-source AI app builder should I choose?

Choose by use case: Dyad for a local, private desktop builder with bring-your-own model keys; bolt.diy for a browser-based Bolt-style experience; Open Lovable to recreate an existing website as editable React; OpenHands for autonomous repository-level engineering; Reflex for Python-first full-stack apps; Appsmith or ToolJet for open-source low-code internal tools; and sandboxd if you are building your own multi-user app-builder product. If you are unsure, test the full create-edit-preview-Git workflow on a small real project before committing.

Is Dyad really open source?

Dyad is an open-core project rather than uniformly permissive. According to research verified against the repository's LICENSE file, everything outside the src/pro/ directory is Apache 2.0 licensed, while content inside src/pro/ is governed by a separate license defined in that directory. In practice this means the core local builder is open source and free to use, but review the license boundaries carefully before embedding or reselling any part of it in a commercial product.

How do I keep AI-generated apps secure when using open-source builders?

Treat generated code as a starting point: review the data model, add authentication and roles, remove unnecessary packages and insecure client-side secrets, and run security scans - some builders like Dyad include built-in security scanning. If you extend your builder or agents with community skills, remember that published research found about 26% of agent skills carry at least one vulnerability, so use a verification layer: SkillzBench is building free security scanning, license checks, quality judgment and sandboxed benchmarking for agent skills, with a 0-100 Trust score, so you only equip skills that are proven safe and effective.

Can I pair SkillzBench with any of these builders?

Yes - that is the intended use. SkillzBench's free MCP tool is being built to connect any agent (Claude Code, Codex, Cursor, Gemini CLI and chat-app agents) to the skills you choose, with per-agent loadouts so each agent only sees what was selected for it. Whether you are building with Dyad locally, bolt.diy in the browser, or generating Appsmith internal tools with a coding agent, verified and benchmarked skills from SkillzBench's directory can automate the rest of the workflow. The platform is pre-launch, so join the waitlist at skillzbench.com for access.