Replit is an AI app builder now, not mainly a code editor. You describe an app in plain English, and Replit Agent plans it, writes the code, tests it in a real browser and publishes it from the same cloud workspace, with hosting, a PostgreSQL database and user sign-in built in. Starter costs $0 with no time limit, Core is $20 per month billed monthly with $20 of usage credits, and Pro starts at $100 per month billed monthly. Is it legit? Yes. Whether it is safe and good value depends on a spending budget you set and on knowing what a rollback cannot undo. Builders who want one place to build, host and ship should pay; anyone who needs a fixed monthly bill should tread carefully.
From browser IDE to an agent that ships whole products
Replit began as a place to write code in a browser, and the old repls were renamed Apps in January 2025. Since Agent 4 arrived in March 2026, the unit of work is a project holding up to seven artifacts: web apps, one mobile app, slide decks, data visualizations, animations and 3D games, sharing one database and backend and published together. Every web app Agent produces is full stack, with API routes and a database added when the idea calls for them. App Storage, Replit Auth or Clerk sign-in and Secrets come built in, while connectors and custom domains start at Core.
How much of that you reach depends on the plan. Starter runs Lite build only, a quick mode for targeted edits, while Full build, Plan Mode, the paid Power and Max modes, connectors and AI Integrations start at Core. Plan Mode is the habit worth forming early: Agent proposes a plan without touching files until you approve it. Tasks also run in isolated copies of the project, so parallel work never lands in the main app until you apply it. The code workspace stays on every plan, and imports arrive from GitHub, Vercel, Figma, Bolt, Lovable and Base44. Lovable, Base44 and Bolt projects bring code and schema but not data records or secrets.
A Replit budget is a hard cap on monthly spending beyond your credits; add a usage alert below it so you hear about runaway Agent work before hosting is suspended.
On Starter you can publish one app, and its link goes down after 30 days unless you publish it again.
Credits are dollars, and effort sets the price
Core and Pro include monthly credits measured in dollars of usage, not messages, while Enterprise buys an annual credit commitment. They pay for Agent work, publishing, network transfer, database storage and AI Integrations, and they reset on your renewal date; only Pro rolls unused credits over, for two months. Agent is billed by effort: a long, complex request spends more than a quick fix, and Replit publishes no price per prompt. Free Mode softens the meter. It is an allowance of no-cost Agent work on Starter, Core and Pro, resetting every five hours with a weekly limit on the paid plans, not a separate free plan and not offered on Enterprise.
Prices shown are what Replit lists for visitors in the United States, in US dollars. Your price can differ by country, currency and local taxes such as VAT.
| Plan | Price, billed monthly | Price, billed annually | Credits included | Builders |
|---|---|---|---|---|
| Starter | $0 | n/a | Daily Agent credits up to a monthly cap (amounts not published) | No collaborators |
| Core | $20 per month, billed monthly | $18 per month, billed annually | $20 a month | Up to 5 |
| Pro, $100 of credits | $100 per month, billed monthly | $90 per month, billed annually | $100 a month | Up to 15, pooled credits |
| Enterprise | Custom | Custom | Annual credit commitment | Unlimited seats |
Prices are per account, not per seat, and Replit says Pro carries no per-user fees. Four larger Pro options follow, with $250, $500, $1,000 and $2,500 of credits a month, the largest at $2,050 per month billed monthly or $2,000 per month billed annually, and the price does not climb in a straight line; our Replit pricing guide lays out every option. By our arithmetic, Pro with $100 of credits costs $80 a month more than Core on monthly billing, for $80 more in monthly credits plus a bigger team and the option of a 28-day database restore window. Usage beyond the credits is billed pay-as-you-go unless you set a budget[1].
That budget is the control that matters most. Replit describes it as a hard cap: once spending reaches it, you get an email and every usage-based service is suspended, hosting and databases included. Set it anyway. Replit itself recommends one so a looping Agent cannot run up a large bill. Traffic is the other surprise, because compute, requests, data transfer, the production database and AI calls made by a published app all come out of the owner's credits, as does Agent work by collaborators you invite.
Rolling back code never restores the production database; live data needs a point-in-time restore, 7 days by default, and Pro can choose up to 28 days.
Data safety: what a rollback restores and what it leaves alone
Any honest answer on whether Replit is safe starts with July 2025. Replit's own account is that SaaStr co-founder Jason Lemkin had the Agent delete database data before Replit separated development from production, and that he restored it fully with Rollback. That same month Replit launched separate development and production databases in beta for new apps and migrated existing ones. Agent now works on a free development database and can only read production through a replica.
The trap that remains is subtler. A checkpoint captures files, conversation context and development data, but the production database is never restored by a rollback[2]. Live data comes back through a point-in-time restore, on by default with a 7-day window that Pro can stretch to 28 days, and that restore replaces the whole database, or from scheduled daily backups (Core and up) if you turned them on. Replit also warns that Agent behaves probabilistically and can get things wrong. Its support team does not debug Agent's code.
Replit says it holds SOC 2 Type II certification, covering its own controls rather than the apps you build, so compliance for a health or payments app rests with you. On training, Replit may use prompts and generated content to improve its models; Pro can opt out[3], and Enterprise content is handled under the Enterprise agreement. Leaving works only in part: code goes out through GitHub sync or a ZIP and the production database is standard PostgreSQL, but Replit Auth accounts, connectors, AI Integrations, App Storage and secrets must be rebuilt elsewhere.
Phones, publishing and team ceilings
Unlike Lovable, which has no native or store output, Replit builds native mobile apps with Expo and React Native. You can preview one on a phone in the free Expo Go app on any plan, while simulator previews need Core. Shipping to the App Store requires your own Apple Developer Program membership, and Replit runs the cloud build and submission, so no Mac is needed. Android is the gap. Google Play is not supported yet. Our Lovable vs Replit comparison weighs that split.
Web publishing is one click on every plan. Apps Agent builds need Autoscale, which scales to zero and bills for compute and requests while serving traffic, or a Reserved VM at a fixed monthly rate, because Static hosting cannot run their backend. Starter gets one published app whose link goes down after 30 days unless you publish it again, with the Made with Replit badge[4]. Core unlocks unlimited published apps, custom domains and badge removal. Private and password-protected publishing has been open to every plan since May 2026.
Growing teams meet the ceilings quickly. Core invites up to 5 builders, Pro up to 15 plus 50 viewers, and Pro runs up to 10 background tasks in parallel against one on Core. SSO, SCIM and audit logs are Enterprise only. Support climbs the same ladder: AI-powered help on Core, direct access to a support engineer on Pro, and a dedicated account manager on self-serve Enterprise.
On Core and Pro, Plan Mode lets Agent propose changes without touching files until you approve. Planning is billed too, but Replit says it uses fewer tokens than code generation.
How our score reads Replit, and what unhappy users report
Replit covers the whole job of an AI app builder on one account: from a plain-language prompt Replit Agent plans, writes the code, tests the app in a real browser and publishes it from the same cloud workspace, and every web app it builds is full stack (frontend, backend, API routes and a database as needed). Since March 2026 one project can hold several artifacts (web apps, a mobile app, slide decks, data visualizations, animations and 3D games) that share one database and backend and are published together. Built in: Replit Database, a managed PostgreSQL with separate development and production databases (since July 2025; Agent reads production only through a replica), point-in-time restore of the production database (7 days by default, up to 28 days on Pro), scheduled daily backups, App Storage, Replit Auth or Clerk Auth, Secrets, AI Integrations, connectors and custom domains bought or connected inside Replit. Hosting comes in four deployment types (Autoscale, Static, Reserved VM, Scheduled) with access set to Public, Password protected, Workspace only or Invite only on every plan since May 2026, a choice of publishing region on paid plans, monitoring, uptime checks on paid plans and a security scan before every publish. Native mobile apps are built with Expo and React Native and submitted to the App Store with your own Apple Developer Program membership, with Replit running the cloud build and no Mac or Xcode needed. Replit Design (the Design Canvas, live for every user since July 2026) turns prompts, Figma files, website URLs or screenshots into frames, and the classic code workspace stays on all plans with a Shell and a Git pane, SSH from VS Code or Cursor on paid plans and imports from GitHub, GitLab, Bitbucket, Vercel, other AI builders and ZIP files. Held back by no Google Play publishing yet (Android apps only in an emulator or Expo Go), Static deployments not fitting apps Agent builds (they need a backend, so Autoscale or Reserved VM), code rollback never restoring the production database, a publishing region that is permanent once an app is published, the Starter plan's single published app going down after 30 days unless republished, and an exit path that carries code, Git history and the standard PostgreSQL database but leaves Replit-managed pieces (Replit Auth accounts, connectors, AI Integrations, App Storage, secrets) to be replaced elsewhere, with running an app outside Replit documented only for Enterprise image exports (early access).
AI is the product itself: Replit Agent builds, tests the app in a real browser and fixes issues in a loop, reviews its own code and triggers a scan on the files it changes. Modes: Free Mode (no-cost Agent work inside an allowance on Starter, Core and Pro, always with intelligent model routing), Power (cost-optimized models, Replit's recommended default) and Max (more capable models for complex, production-grade work), each with Auto routing per task or, on Core and Pro, a manually chosen model, plus a per-model Effort control and Compare models (beta), which runs one prompt on several models as separate agents. Plan Mode (Core and up) reads the project and changes no files until you approve a plan; tasks run in parallel in isolated copies of the project on a shared Kanban board and nothing touches the main project until you apply it; Routines schedule Agent runs with a per-run budget; the Monitoring Agent (October 2026) investigates alerts and validates fixes, and the Security Agent reviews code on paid plans. Replit says its intelligent model routing gives the same output quality at lower cost than the previous Max Mode in its own testing (August 2026). Inside apps, AI Integrations give models from OpenAI, Anthropic, Google Gemini, Grok and OpenRouter without your own API keys, billed at the provider's public API price from Replit credits. Held back by cost uncertainty: Agent is billed by effort, so a complex request costs more than a simple one, and Replit publishes no current price per request (only dated examples, such as simple changes typically under $0.25 in its June 2025 post); Plan Mode interactions and App Testing are billed too; Free Mode's allowance resets every five hours with a weekly limit on Core and Pro and Replit publishes no allowance numbers for Starter or for Pro beyond «Even more»; Starter gets Lite build only, with Full build, Plan Mode, Power and Max on Core or higher; Replit itself warns that Agent's behaviour is probabilistic and may make mistakes; and Replit may use prompts and generated content to improve its models, with an opt-out only on Pro and Enterprise content handled under the Enterprise agreement.
Replit integrates at several levels. Connectors sign in once with OAuth while Replit manages the credentials centrally (launched in autumn 2025, each powered by MCP; the full set on Core and Pro since April 2026, managed workspace-wide by the account admin), covering Google Workspace, Microsoft 365, GitHub, GitLab, Jira, Linear, Slack, Salesforce, HubSpot, Zendesk, Stripe, Square, Shopify, Plaid, RevenueCat, Airtable, Notion, BigQuery and Snowflake; Replit's changelog (June 2026) describes one catalogue spanning connectors, MCP servers and Git providers. MCP works both ways: Agent connects to one-click MCP servers or any remote HTTPS MCP server, with MCP traffic passing Replit's security scanner, and the Replit MCP Server (public beta) lets ChatGPT, Claude, Slack or any MCP client create, find, update and publish Replit apps; Replit is also an app in ChatGPT (December 2025, not in the EU) and a connector in Claude (June 2026). AI Integrations put OpenAI, Anthropic, Google Gemini, Grok and OpenRouter models into apps without your own keys. For selling, Agent sets up Stripe (paid plans, starting in a sandbox, going live through the Replit Integrated Payments app on the Stripe Marketplace with no key pasting), with Whop for digital products, Shopify on every plan and RevenueCat for native mobile subscriptions; around them sit two-way GitHub sync in the Git pane, ZIP download, a Slack bot, Figma import and imports from other builders. Connecting tools and connector authentication or data reads use no credits. Held back by gates: Starter has no third-party connectors (Shopify excepted), no Replit AI Integrations (its own API key still works) and no Stripe integration; custom connectors (beta) need Pro; warehouse connectors (BigQuery, Snowflake, Microsoft Fabric), custom OAuth apps for connectors and self-hosted Git providers need Enterprise; AI Integrations stay off on Pro and Enterprise until an admin enables them; and third-party AI and Agent services are deducted from Replit credits at the provider's rate.
Easy to start: describe the app in the prompt box (or by voice, file attachments or a Figma link), Agent picks a stack, builds a full-stack app and publishes it with one click, with hosting, SSL and infrastructure handled by Replit. Free Mode lets everyday chatting and building run without spending credits while the allowance lasts and asks for confirmation (Allow once) before any paid action; Plan Mode lets a non-coder approve a plan before anything changes; the Visual Editor makes simple text, colour and spacing edits without credits; checkpoints capture code, Agent context and, as an option, the development database for rollback; the Usage page and the chat header show costs as work runs; and the Replit mobile app lets you build and watch deployments from a phone. Held back by how much a non-coder has to understand: modes, models and effort levels that change the cost of a request; every hosting and database item metered from the owner's credits; usage beyond the credits billed pay-as-you-go unless a budget is set, and a budget that, once reached, suspends all usage-based services including hosting and databases, with deployments that do not restart on their own after a billing hold; a separate production database whose schema changes are applied at publish with data-loss warnings to read; Static deployments that do not fit apps Agent builds; a publishing region that cannot be changed after publishing; and, on Starter, Lite build only and one published app that goes down after 30 days unless republished.
Replit is priced per account, not per seat: Core invites up to 5 builders with no separate team subscription, and Replit says Pro allows up to 15 builders with pooled credits and no per-user fees (November 2026). Starter is free with no time limit (daily Agent credits up to a monthly cap and monthly cloud credits, amounts not published, Free Mode, Lite build and one published app). Core costs $20 per month per account billed monthly or $18 per month billed annually ($216 a year billed annually, by our arithmetic) and includes $20 of credits a month; Pro starts at $100 per month per account billed monthly or $90 per month billed annually for $100 of credits a month ($1,080 a year billed annually, by our arithmetic), and its larger options run up to $2,500 of credits a month for $2,050 per month billed monthly or $2,000 per month billed annually, buying credits below face value (18% below at that option on monthly billing, by our arithmetic). Credits are dollars of usage that cover Agent, publishing, databases and AI Integrations; Free Mode adds no-cost Agent work on Starter, Core and Pro (Core's card: up to 30 hours of chat on Free Mode); Pro's unused credits roll over for two months; credit packs on Core and Pro run from $100 for $100 of credits to $950 for $1,000. Held back by Agent work priced by effort with no fixed price per request; usage beyond the credits billed pay-as-you-go unless a budget is set, and a budget that suspends hosting and databases when reached; every hosting and database item metered (Autoscale at $0.60 per million compute units and $0.40 per million requests, Reserved VMs from $15 per month for a shared 0.5 vCPU, database compute at $0.16 per compute hour and storage at $0.35 per GiB per month); a published app's visitors spending the owner's credits; Core credits resetting each month with no rollover; credit packs expiring six months after purchase; Plan Mode billed; annual billing only up to 10% cheaper per month (Core and Pro with $100 of credits, less on larger Pro options); the AI-training opt-out only on Pro; and Agent, AI and other usage charges plus domain fees being non-refundable on every plan (the Terms allow full refunds of recent Starter and Core subscription payments, and Pro refunds are handled case by case by Replit Support).
Support is tiered and documented in a support policy: the Replit Community Hub for every builder, AI-powered support for technical and billing questions on Core with billing questions escalated to the support team, prioritized direct access to a support engineer on Pro and Enterprise in addition to AI support, and dedicated premium support with an account manager on Enterprise; since November 2026 Agent checks active incidents and the docs before a ticket, and on Pro it can file the ticket with debugging details. On reliability and security, as Replit describes them: Replit says it holds SOC 2 Type II (announced August 2025), complies with GDPR and is working toward ISO 27001; every customer gets an isolated Google Cloud project, Cloud Armor DDoS and WAF protection covers every deployment, the Package Firewall has been on by default since June 2026, a security scan runs before every publish, the production database has point-in-time restore (7 days by default, up to 28 days on Pro) and optional scheduled daily backups, and Replit's deployments page advertises 99.9% uptime for Reserved VMs (read November 2026). Held back by no phone or live-chat support and no published uptime SLA for self-serve plans, Pro support through web tickets started by the customer, support that does not debug your application code, a budget or unpaid invoice that suspends hosting and databases (deployments must be published again after a billing hold), the app's own compliance (GDPR, HIPAA, PCI DSS) left to the builder with no BAA published, a bank dispute filed before contacting Support possibly locking the account, and incidents as Replit describes them: in July 2025 Agent deleted database data of SaaStr co-founder Jason Lemkin before development and production databases were separated (restored in full with Rollback, Replit says), a July 2025 billing error inflated checkpoint charges for some paying users (refunded or credited automatically, Replit says), and widely disclosed vulnerabilities in October 2026 required some Node.js and PHP apps to be republished.
Mean of each platform's displayed rating with at least 30 reviews, weighted by review count (read November 2026): Trustpilot 2.8 from 1,562 reviews, G2 4.4 (its review count differs between indexed snapshots, every one above 30), Capterra 4.4 from 157 reviews and Product Hunt 4.6 from 52 reviews give about 3.2, so the axis rests mostly on Trustpilot. Trustpilot's negative reviews cluster around costs and credits, Agent loops, support and locked accounts; that is a map of complaints, not a measure of all users. The review sites G2, Capterra and Product Hunt are far warmer. No ad-hoc adjustment is applied.
docs.replit.com is deep and specific: billing pages on the plans, credits, managing spend (usage alerts, budgets, credit packs), usage-based billing and the August 2026 cloud price change with old and new rates; feature pages on Agent modes, Free Mode, Plan Mode, checkpoints and rollbacks, artifacts, Replit Design, publishing and deployment types, project geography, custom domains and domain purchasing, the development and production databases and data recovery, App Storage, Replit Auth and Clerk, connectors, MCP and AI Integrations, mobile apps from Expo Go to TestFlight, security scans and the shared-responsibility model; Enterprise and team administration; a help FAQ; dated legal pages (Terms of Service August 2026, Privacy Policy October 2026, sub-processors August 2026) and a support policy; a product changelog dated entry by entry; plus Replit Learn, a free course platform on building apps with AI. Held back by most help pages being undated and by older pages that were not updated and contradict the pricing page or newer dated entries (an older annual Core figure in the pricing help, a one-month rollover and an older credit range on the Pro product page, a no-rollover line on the deployment pricing page, access controls for Pro and Enterprise only and United States-only hosting on an older deployments page, legacy Teams seat and cancellation pages), and by Replit's comparison pages being a competitor's claims, which leaves the pricing page, the legal pages, the billing docs and the dated changelog as the reliable sources.
Weighted across the axes above (weights reflect what actually drives the decision in this category). How we score →
The score rewards how much of the path from prompt to live app Replit covers in one account: agent, database, sign-in, hosting, iOS publishing and security scans, with a real exit route for the code. The meter holds it back. A bill that grows with effort and traffic is hard to forecast, the longer restore window and the training opt-out start at Pro, and the Android gap matters to phone-first builders. That balance puts Replit at 3.9/5, the third product we have scored among AI app builders, with no rank implied.
Ratings split sharply. Review sites where business users describe daily work read far kinder than Trustpilot, where the complaints we read cluster around costs, credits and unexpected charges, Agent loops and bugs, support, locked accounts, canceling, refunds and lost data. Treat that as a map of where things go wrong, not a measure of how often they do. Most of those themes lead back to the budget and data points above.
Who should pay, and how far up the ladder
Core fits a solo builder or a small group shipping a few web apps, internal tools or an iOS prototype who wants Full build, Plan Mode, connectors and a custom domain. Pro earns its place once you need more than five builders, pooled credits, the option of a 28-day restore window, the training opt-out or faster answers from a human. Enterprise is the step for SSO and audit logs. There is no trial of Core or Pro, so start with a Starter project in Free Mode before paying.
Look elsewhere if Google Play is a launch requirement, if you need a bill that never moves, or if you handle regulated health data, since Replit publishes no HIPAA business associate agreement we could find. Base44 produces store-ready wrapper apps, your published web app in a web view, for both Apple and Google Play from its Builder plan, as our Base44 review explains. Replit's Terms of Service allow full refunds of Starter and Core subscription payments made within 30 days of purchase, Pro refunds are handled case by case by Support, and usage, Agent and domain charges are non-refundable.
Frequently asked questions
Can I pick which AI model Replit Agent uses?+
On Core and Pro you can choose an available model manually in Power or Max mode, or let Auto route the work. Free Mode always routes automatically, and Fast model variants need Pro or Enterprise.
Can I pause a Replit subscription instead of canceling?+
Some subscribers can pause once for one month. You pay nothing and get no monthly credits, published projects stay live, and Agent and hosting usage is still billed pay-as-you-go.
What happens to my apps if I cancel a paid Replit plan?+
You keep plan features until the billing period ends, then the account moves to Starter. Published apps may go offline, custom domain connections are removed and storage is reduced.
Will Replit delete a free account I stop using?+
A free account with no login for one year counts as inactive, and Replit may terminate it and delete its data and apps. Paid accounts are exempt.
The bottom line
Replit has turned into an agent that takes an idea to a hosted, database-backed app, and to an iOS app if you bring an Apple Developer account, without leaving the browser. That breadth is real, and our score credits it. Its price is a dollar meter driven by effort and traffic, so a budget is not optional, and its production data is something a code rollback cannot bring back. Pay for Core if you build alone or with up to five builders and keep the budget tight; move to Pro when you need more builders, pooled credits, rollover or the longer restore window.
- [1] Replit plans and pricing (2026-11) https://replit.com/pricing
- [2] Replit docs: Checkpoints and Rollbacks (2026-11) https://docs.replit.com/features/version-control/checkpoints-and-rollbacks
- [3] Replit docs: Model improvement (2026-11) https://docs.replit.com/legal-and-security-info/model-improvement
- [4] Starter Plan (2026-11) https://docs.replit.com/billing/plans