The short answer
Replit is a browser-based platform where Replit Agent turns plain-language requests into working web and mobile apps, dashboards, designs and slides, then hosts and publishes them with a database included. Agent 4 launched on 13 March 2026. You choose Free Mode, Power Mode or Max Mode, with models such as Claude Sonnet 5.5 and GPT-6.1 Sol. The free Starter plan allows one published app, which goes down after 30 days.
| Made by | Replit, United States |
|---|---|
| Free plan | Starter: daily Agent credits, Lite build, one published app for 30 days |
| Models | Auto routing, plus Claude Opus 5.5, Sonnet 5.5, Haiku 5.5, GPT-6 Sol and GPT-6.1 Sol by mode and plan |
| Best at | Full apps with backend, database and hosting in one place |
| Shows sources | No, it is an app builder rather than a research tool |
| Team plans | Pro for up to 15 collaborators, Enterprise with SSO and SCIM |
| Your data | Pro can opt out of model improvement. Enterprise content is not used for training. No opt-out is documented for Starter or Core. |
What Replit is in late 2026
Replit began as an online coding environment and is now mostly an AI builder. You describe what you want and Replit Agent plans it, writes the code, sets up the database and publishes the result, all in the browser. Agent 4, released on 13 March 2026, widened what it makes: web and mobile apps, data dashboards, AI-powered tools, designs and slides. It added a Design Canvas for mockups before code, Plan mode to map out a build first, parallel tasks and collaboration with teammates in the same project.
The way you pay for that work changed in August 2026. You now pick a mode. Free Mode lets you chat and build without spending Power Mode or Max Mode credits, with limits that depend on your plan (you can see what is left under Settings, then Usage). Power Mode and Max Mode spend credits and give you stronger models. Inside Power or Max you can select Auto, which since 28 August 2026 routes each part of the job to a model and effort level that fits its difficulty. You can also choose a model yourself. Recent additions include Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna Fast (25 September), GPT-6.1 Sol in Max Mode and Claude Sonnet 5.5 in Power Mode (2 October), and Claude Haiku 5.5 in Power Mode for Core and Pro (16 October).
You can use Replit in the browser, in desktop apps for macOS and Windows, and in mobile apps for iOS and Android.
Where Replit is genuinely strong
Apps that need a backend
Replit's real advantage is that the boring parts are already there: database, storage, logins and publishing. A team lead can ask for an internal request tracker with a login and an admin view and get something people can actually use the same afternoon, without opening an account anywhere else.
Starting with a conversation
Since 21 August 2026 you can begin in a private Conversation to ask questions, think through an idea or gather context, and turn it into a Project when it grows. Your context and files carry over. This suits people who do not yet know exactly what they want to build.
Repeated work on a schedule
Routines let you schedule repeated work inside a Conversation, such as a weekly report, with results returned to the same thread. They need Power or Max Mode and a budget for each run. Since September you can also ask for charts and reports directly in a conversation, after Replit announced its acquisition of the charting company Atta.
Connecting your work tools
On Core and above, Agent can use connectors to pull data into the chat from tools such as BigQuery, Linear, Slack and Notion. You can import Skills from a public GitHub repository so Agent follows your team's instructions, and since 28 August 2026 Growth Skills help find first customers with partners such as Apollo, Clay, Stripe and PostHog.
Where it falls short
- The free plan is temporary. Starter gives you Lite build only and one published app, which goes down after 30 days and carries a Made with Replit badge. Plan Mode, connectors and Replit AI Integrations need Core.
- Costs are hard to predict. Power and Max spend credits by the job, and an open-ended request can use more than you expect. Check Settings, then Usage, before and after a big build.
- Privacy depends on the plan. Only Pro has a documented opt-out from model improvement, and Enterprise is excluded by default. See the privacy section below.
- Polish takes extra rounds. Replit is built for apps that work. If the main goal is a striking interface, v0 often gets there in fewer steps.
How to get much better results from Replit Agent
- Use Plan mode first (Core and above). Ask Agent to lay out the screens, data and steps, read the plan and correct it before any code is written. Fixing a plan costs far less than fixing an app.
- Explore in Free Mode, build in Power. Use Free Mode for questions and rough ideas, then switch to Power Mode with Auto for the actual build. Save Max Mode for hard problems.
- Describe the data. Say what the app stores, who can see what and which actions change it. Most broken builder apps come from unclear data rather than unclear screens.
- Steer instead of starting over. With Steer, which is the default, you can send follow-ups while Agent works to add context to the current task. If you prefer to wait, switch to Queue under Settings, then Personalization.
- Run a security scan before you share. Replit offers a Level 3 scan that combines dependency checks, static analysis and an external test of the live Preview. Use it on anything that handles logins or customer data.
Our Replit lessons take you through a first app, and the starter kits include ready-made briefs you can paste into Agent.
Privacy settings to check today
Replit's documentation says Pro users can opt out of their content being used to develop and improve its models: open Settings, go to Personalization and turn off Model Improvement. The opt-out applies to new content only. For Enterprise users, Replit does not use customer content to train its models. The documentation does not describe an opt-out for Starter or Core, and Replit's privacy policy, updated on 21 October 2026, lists improving code generation as a legitimate interest for using personal data. If you are building with customer data, work on Pro or Enterprise, or keep real data out until you have checked with your company.
Which Replit plan suits whom
Starter suits anyone trying Replit for the first time or building a throwaway demo. Core suits individuals who want apps that stay online, with Full build, unlimited published apps, connectors, Plan Mode and no badge. Pro suits small teams and anyone who needs the training opt-out, with up to 15 collaborators, database rollback of up to 28 days and premium support. Enterprise adds SSO, SCIM, audit logs, model controls for each Workspace and, since 16 October 2026, options to run Replit or published apps in your own cloud, offered in private preview and early access. Check Replit's pricing page for current prices.
Replit is not the best pick if you only need a landing page or if your developers want to work in their own editor from day one. To see how it stacks up, read Lovable vs Replit or try our comparison tool.
Prompts to copy
Plan before building
I want to build [app] for [who uses it]. Before writing any code, give me a plan: the screens, what data the app stores, who can see and change what, and the order you would build it in. Ask me up to five questions about anything unclear.
You get: A build plan you can correct before spending credits
Internal tracker with logins
Build a [request or issue] tracker for our [team]. People sign in, submit a request with [fields], and see only their own requests. Admins see everything and can change status and add notes. Send the requester an email when the status changes. Use the built-in database.
You get: A working internal tool with roles and storage
Weekly report as a Routine
Every Monday at 9 a.m., pull last week's [data] from [connected source], summarise the three biggest changes in plain English, add one chart, and post the result in this conversation. Keep the summary under 150 words.
You get: A scheduled report that arrives without you asking
Customer-facing form
Build a booking form for [business] where customers pick a [service], a date and a time slot from [hours], and enter a name and email. Block double bookings. Give me an admin page to see and cancel bookings, and show me how to test it.
You get: A bookable app with a test plan
Explain before you change
Before you change anything, explain in plain English how [feature] works in this app today, which files it touches and what could break if we change it. Then suggest the smallest change that achieves [goal].
You get: A safer change with the risks named
Pre-launch check
Review this app before I share it with [audience]. Check logins and permissions, what happens with empty or wrong input, how it looks on a phone and anything that exposes data it should not. List the issues by severity and fix the serious ones.
You get: A prioritised list of fixes before anyone else sees it