- Why the second product nobody sees costs so much
- What changed this year: the coding agent became the builder
- Three forks that decide the shortlist
- The eight builders
- What internal tool builders cost
- An afternoon test with your own schema
- Mistakes that turn internal apps into a liability
- My recommendation by type of team
An internal tool builder turns your databases and APIs into admin panels, dashboards and forms far faster than building them from scratch. Choose by who'll maintain the result: open-source builders you can self-host, code-first frameworks for developers, or AI-led platforms that put governance around apps generated by coding agents.
The best internal tool builder is the one that matches who'll still be maintaining the app in a year: an open-source builder you can host yourself, a code-first framework your developers own, or an AI-led platform with governance wrapped around it. Pick that route first and the shortlist shrinks from eight to two before you've opened a single demo.
This guide is for developers and technical founders who keep getting asked for one more admin screen. The tools are further down, one line each. The thinking you need to choose between them comes first.
Why the second product nobody sees costs so much
Nobody plans to spend a sprint on an internal tool. It starts with support asking for a way to refund an order without a database query. Then sales wants to edit account limits. Then finance needs an export every Monday. Each request is an afternoon. Together they become a second product with its own logins, permissions, bugs and on-call pages, and not one customer ever sees it.
Internal tool builders exist to make those screens cheap. They connect to the Postgres, MySQL, REST and GraphQL sources you already run, hand you tables, forms and buttons that read and write that data, and look after who's allowed to press what. Where they differ is who builds, where the app runs, and how much of the result you can read as code when something breaks at 2am.
What changed this year: the coding agent became the builder
A year or two ago the pitch was drag and drop. Now nearly every vendor in this space leads with AI coding agents. The biggest name offers an MCP server and a CLI so an agent can build an app and deploy it into its governed environment. ToolJet ships its own MCP server for the same job. Superblocks goes a step further and sells governance as the product, assuming your staff will build with whichever coding agent they already like.
That's useful, but it moves the hard question. Generating a working screen from a prompt is the easy half. The hard half is what happens on day two: can a developer read and change what the agent produced, does it respect the permissions you set, and does an app built by someone in operations go through the same review as one built by an engineer? Ask every vendor to show you an AI-built app after a human has edited it, not the shiny first draft.
Three forks that decide the shortlist
Where does it run? Open source and self-hosted builders keep data on your own servers, which matters if you're in health, finance or anything with a data residency clause. The trade is that your team owns upgrades, backups and uptime. Hosted builders take that off your plate, and most of the open-source ones also sell a hosted version if you change your mind.
Canvas or code? A visual canvas is quicker for a CRUD screen and lets non-developers tweak a label without a pull request. Code first tools give you version control, code review and an app that lives in your repository like everything else. If your engineers already resent tools they can't diff, don't buy them a canvas.
Who builds? If it's three developers, governance is a nice-to-have. If it's thirty people across operations, sales and finance, it's the whole point. AI-led with governance is the route for that second company: central control over data connections, single sign-on, role-based access and audit logs, so IT knows what reached production and who put it there.
The eight builders
Appsmith
An open-source low-code platform for internal apps built on databases, APIs and LLMs, which you can host yourself when internal data mustn't leave the building.
Budibase
Leans towards whole internal processes rather than single screens, bundling apps, event-driven automations, data tables and a service portal in one open-source package.
Refine
A React framework instead of a canvas, turning REST APIs or Supabase projects into internal apps that sit in your own codebase.
Retool
The best-known name in the category, now centred on prompting an app into existence and deploying it inside a governed environment, with an MCP server and CLI for agents.
Superblocks
Made for IT and platform teams who want a single governed route to production for the agents and internal apps staff build with their own coding tools.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
ToolJet
An AI app builder for internal tools that comes with its own database, AI agents and an MCP server, and runs on your own infrastructure if you'd rather.
UI Bakery
Fits admin panels, CRUD screens and forms, including apps assembled from MCP tools, with SSO, role-based access and audit logs that security teams look for.
Windmill
Code-first and open source: scripts in Python, TypeScript, Go, PHP, Bash or SQL become workflows, generated interfaces and small apps.
What internal tool builders cost
Pricing in this category clusters into three shapes. Most builders have a free tier or a free self-hosted edition, usually capped at a handful of users or builders, which is plenty for a pilot. Paid cloud plans tend to charge per builder or per user, landing somewhere around $10 to $25 a month each at the entry level. A few sell a flat team plan instead, and those start nearer $100 to $300 a month. Enterprise contracts with SSO, audit logs and support commitments are quoted, and for a hundred users they can run into thousands a month.
Two things catch people out. First, "per builder" and "per user" are very different bills once a tool spreads beyond engineering, so count the people who'll only view and click, not just the people who'll build. Second, some builders now price on AI credits, so a team that leans on prompt-to-app generation pays for usage on top of seats. Self-hosting looks free, but it isn't: someone's time goes on patching it.
An afternoon test with your own schema
Pick the internal screen you've been dodging. A good one: support looks up an order by email address, sees the last five orders, and can issue a refund, but only people in the support lead role can press the refund button. Build it twice, once in each of your two strongest candidates, against a copy of your real database. Sample data hides the thing builders differ on most, which is how they cope with your schema, your naming and your odd joins.
Time yourself to the first working lookup. Then add the role rule and try to break it by logging in as an ordinary agent. Then make one change a week later would need, like adding a refund reason field, and see whether that's a two-minute edit or a rebuild. Finally, hand the app to a developer who wasn't there and ask them to explain what it does in five minutes. If they can't, that's your next two years of maintenance.
If the refund screen also needs to fire a webhook to your billing system, test that hop on its own first with one of these webhook testing tools, so you know whether a failure is the builder or the endpoint.
Mistakes that turn internal apps into a liability
- Giving the builder a database user with full write access because it's quicker. Create a scoped user per app and let the builder's own permissions sit on top, not instead.
- Choosing on the prettiest demo. Every builder looks good on a to-do list; yours has fifteen tables and a column called status2.
- Letting AI-generated apps skip review because they were fast to make. Speed of building has nothing to do with safety of running.
- Forgetting the exit. Ask how you'd get an app out if you left. Code in your repository is portable; a proprietary canvas usually isn't.
- Pricing for today's five builders when the plan is to let thirty people in by next year.
My recommendation by type of team
If you're a small developer team that hates anything you can't put in Git, start with a code-first framework or a code-first script platform. If your data can't leave your own servers, shortlist the open-source builders and budget real hours for running them. If lots of people outside engineering are going to build, and some of them will use coding agents whether you approve or not, a governed AI platform is worth its higher price because the control is what you're buying.
Whichever way you lean, ship one screen to a small group before you move anything else across. If you're leaving an existing platform, my list of Retool alternatives covers that switch, and for customer-facing apps rather than internal ones, these no-code app builders are the better fit.
Do you make an internal tool builder or low-code app platform? Tool makers can ask to be considered for a featured place on this list.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.


