Bubble Review 2026: The Deepest Visual Builder Now Has an AI Agent — and a Meter You Need to Understand
Bubble is the long-running no-code platform for building real web apps with a database, workflows, privacy rules and hosting in one editor. In 2026 it added an AI Agent that plans and edits inside that editor, a native mobile builder still in beta, and an MCP server that has been announced but not confirmed live. Plans run $0, $29, $119 and $349 for web, plus a workload-unit meter. This review checks all of that against Bubble’s own pricing, manual and security pages, and sets it beside Trustpilot, Capterra and what Bubble’s forum members say about cost.
| Free plan | Yes -- build and test, cannot go live |
| Paid | From $29/mo annual ($32 monthly), web-only Starter |
| AI Agent | Beta since Oct 16, 2025 |
| Capterra | 4.6/5 (334 reviews) |
| Trustpilot | 2.5/5 (153 reviews) |
Bubble is still the deepest all-in-one visual app builder you can buy — but you should price the app before you build it, not after.
“Bubble gives you a database, a workflow engine, privacy rules, web and (in beta) native mobile in one editor, and in 2026 it finally added an AI Agent that works inside that editor. The cost of that depth shows up in two places: a usage meter called workload units that you can’t fully predict in advance, and a platform you can’t take your app out of. Both are manageable if you know about them on day one.”
This is a VERIFIED EDITORIAL REVIEW: Bubble’s own pricing page, manual, security page, blog and AMA recaps, Trustpilot, Capterra, and two recent threads on Bubble’s official forum, all read on October 3, 2026. I did not build an app in Bubble for this article, so I won’t pretend to have timed the editor or measured the agent.
If you’re a non-technical founder who wants a real, multi-user product with logins, payments and a back office, Bubble is one of very few tools where you aren’t assembling five services. If you want a polished prototype by tonight, a prompt-to-code tool will get you a nicer-looking first screen faster. If you care about owning your code, this is the wrong tool: I found nothing in Bubble’s manual describing an app source-code export.
One editor, one database, one hosted runtime — and now an agent that edits all three.
Bubble’s homepage calls it “the only no-code AI app builder.” Strip the slogan and the product is a visual editor with five parts you actually touch every day: a design canvas (elements and responsive layout), a database (data types, option sets, privacy rules), a workflow engine (events that trigger actions, in the browser or on the server), a plugin and API Connector layer, and hosting on shared AWS infrastructure, with dedicated clusters for Enterprise. A native iOS/Android editor sits on the same backend. Bubble’s own docs say workload from web and mobile is combined because both share one database.
The line that matters for budgeting is the last one. The plan you pay for is attached to one app (the Settings tab shows “the current subscription plan for the app”), and the meter is workload.
A decade-old category leader, a $100M round in 2021, and an AI-builder market that moved around it.
Led by Insight Partners, announced Jul 27, 2021 (TechCrunch, Bubble blog). Five years old: not a 2026 funding signal.
Company claim at the Series A, per press coverage of the announcement. Registered accounts, not paying customers.
capterra.co.uk listing, 90% recommend. Reviews are self-selected buyers, not market share.
66% of 153 reviews are one-star; many cite billing and cancellation. See Real Users.
| Date | Event | Source |
|---|---|---|
| Jul 27, 2021 | $100M Series A led by Insight Partners | TechCrunch / Bubble blog |
| Oct 16, 2025 | Bubble AI Agent launched in beta to all users globally; billed as the first AI agent for visual development | Business Wire / SiliconANGLE |
| Mar 2026 | Agent rollout opened in phases: blank apps first, then agencies and ambassadors, then paying apps, then broader over about three months | Bubble March 2026 AMA recap |
| Jul 2026 | Bubble blog describes its native mobile builder as “in beta” | Bubble blog |
| Sep 25, 2026 | AMA recap: agent now on Anthropic’s Claude Sonnet 5, MCP support “coming to beta”, BubbleGo testing app being retired, mobile plugin editor in public beta | Bubble September 2026 AMA recap |
Builders who learn Bubble love what it can do. The loudest complaints are about money.
Reviewers describe rapid MVP building (“weeks” instead of months), strong security features and a helpful community, and flag a steep learning curve, Safari signup issues and complex workflow management as projects grow. Category scores shown: ease of use 4.3, support 4.3, value 4.5, features 4.5.
TrustScore 2.5 from 153 reviews (66% one-star, 22% five-star). A one-star review dated Aug 25, 2026: “auto-pay forever after stopping using the app” with no visible cancellation button. A five-star: “Their customer support and timeliness in fixing bugs is superb!” Trustpilot notes the company replied to only 9% of negative reviews.
In a 33-post thread on staying competitive in the AI era, one builder wrote: “a person has to pay per app, $32 just to build in bubble. And then would have to upgrade to the growth plan $134 to get access to critical features like branching.” Others argued “Bubble chooses the stack for you” is becoming a “phantom moat” now that AI can scaffold stacks.
Free to build, $29 to launch, and a usage meter on top.
Verified live on bubble.io/pricing and in Bubble’s manual on October 3, 2026. The page shows three product types: Web only, Mobile only and Web + Mobile. Below are the web-only prices (monthly billing / annual billing, per month), straight from the manual’s pricing article.
Free
Build and test only: 50K workload units, development version, API Connector, one app editor, 6 hours of server logs. No live deployment or app-store publishing.
Starter
175K workload units, live site, custom domain, custom branding, recurring workflows, basic version control, 2 days of server logs. Web+Mobile is $59 / $69.
Growth
250K workload units, 2 app editors with 10 custom branches, premium version control, two-factor authentication, 14 days of logs. Web+Mobile is $209 / $249.
Team
500K workload units, 5 app editors with 25 custom branches, sub apps, 20 days of logs. Web+Mobile is $549 / $649.
Enterprise
Customizable workload, hosting location and server configuration, dedicated support, invoice or ACH billing, SSO provisioning (per Bubble’s security page).
Overage and storage
Overage for apps on a workload tier, per a third-party breakdown; file storage add-on is $3 per 100 GB per month on the pricing page. Flexible overages can be switched off.
| If you are... | What verified pricing means | Cheapest sensible path |
|---|---|---|
| Learning Bubble or testing an idea | The free plan is real but cannot go live | Free ($0) |
| Launching one web app with real users | Starter includes domain, branding and backend workflows | Starter, $29/mo annual or $32 monthly |
| Need branching and safer releases | Branches begin at Growth, 10 branches and 2 editors | Growth, $119 annual / $134 monthly |
| Want iOS/Android as well | Web+Mobile Starter is $59 annual / $69 monthly, with 5 build submissions a month after the first 3 months | Web+Mobile Starter, accepting the mobile beta label |
| Run several apps | The plan is per app, so each production app carries its own subscription | Budget one plan per app |
Pricing verified Oct 3, 2026. No affiliate link or discount is implied. I found no AI credit allowance on the pricing page or in the AI Agent docs; whether the agent has its own limits is something to confirm in your own account.
Where Bubble fits in 2026.
Bubble is built for exactly this. Budget a month to learn it and one plan per app.
A prompt-to-code builder is faster to a first screen. Compare Lovable, Bolt.new or Base44.
Bubble is the wrong shape. Look at Replit or v0, which produce code you can take with you.
Model workload units first. The forum suggests lean apps are cheap and unoptimized ones are not.
It exists, but Bubble itself calls the mobile builder beta. Treat it as promising, not settled.
The agent proposes a plan you can read, then edits the same pages, data and workflows you would edit by hand.
Bubble’s manual describes the AI Agent as an in-editor assistant that “understands the structure of your app: pages, views, workflows, data types, and privacy rules,” so you can refer to components by name. You can type or speak (voice works in Chrome, Brave, Firefox, Safari and Edge), select elements first for context, and keep separate chat threads per task. For a new app it generates pages, a database, workflows and sample data; for an existing app it adds pages, changes workflows, structures data, connects APIs and adjusts styling. It can also explain how an app works.
The detail I like most is the plan step. In Bubble’s own example, a prompt like “Create a page for signing up and logging in” returns a numbered plan (page size, background, a tab state called auth_tab, login and signup groups with their buttons and navigation) and asks “Would you like me to apply these edits?” with Cancel and Approve plan buttons. It is a diff you can read before anything changes.
| Capability | What Bubble documents | What I could not verify |
|---|---|---|
| Generate an app from a prompt | Pages, database, workflows and sample data | Quality of output on complex apps |
| Edit an existing app | Add pages, edit workflows, structure data, connect APIs, restyle | How it handles large legacy apps |
| Permissions | Same permissions as the user who started the chat; read-only collaborators are declined | Whether there are per-seat or per-plan agent limits |
| Model | Sep 2026 AMA: runs on Anthropic’s Claude Sonnet 5; can create API Connector calls and animations | Which model each plan uses; credit or usage model |
| Reach beyond the editor | MCP support “coming to beta” so tools like Claude or ChatGPT can edit apps | Whether it has shipped; it was not live in what I read |
- The plan-then-approve loop is readable and reversible
- Works on the same primitives (privacy rules, workflows) that you can audit by hand
- Founder says roughly half of user problems stem from structurally unsound apps from the start; the agent aims to catch that early
- Agent usage and limits are not documented where I looked
- It can build quickly on a foundation you may not understand; privacy rules still need your review
- Forum builders argue the platform primitives, not the model, are the real ceiling
Describe it, approve the plan, tune it visually, branch, deploy.
The documented loop mixes AI and manual editing. You describe the app, approve the agent’s plan, then refine on the canvas, in the data tab and in workflows (which Bubble defines as an event followed by actions). On Starter you get basic version control; from Growth you get branches, merge review and conflict resolution, with a hotfix path and savepoints in the docs.
See it in Bubble’s own words
Official video from Bubble’s own YouTube channel (@BubbleIO, channel UCPDgLFs6WYk-qj16lEyFFlw, channel name “Bubble”). It is the Oct 16, 2025 Bubble Tour London opening keynote introducing the AI Agent, about 43 minutes. It is a launch keynote, so treat claims as vendor claims.
Backend workflows are the quiet engine; mobile is the newest and least settled piece.
Backend (server-side) workflows, database trigger events, a scheduler and a Workflow API let a Bubble app do work without a user on the page. Those are also where workload accumulates: in Bubble’s documented rates a server-side workflow action costs 0.6 WU and a scheduler addition costs 0.1 WU. The September 2026 AMA says branching (if/else) and loops are coming to backend workflows, which tells you they were not there as first-class constructs when I read it.
On mobile, Bubble lets you build native iOS and Android views in the same editor, sharing the database, with simulated device testing on the free plan and dev builds on paid plans. The same recap says the BubbleGo testing app is being discontinued because of App Store policy restrictions, and Bubble’s own blog says the mobile builder “is in beta and keeps improving.”
Plugins, an API Connector, and an MCP server that is still on the roadmap.
Bubble integrates through a plugin marketplace, the API Connector (any REST API), a SQL connector, authentication plugins (Google, Facebook, LinkedIn, Slack and others are in the docs) and its own Data and Workflow APIs. A third-party review cites 6,500+ integrations; I could not verify that number on Bubble’s site, so treat it as an estimate. Quality varies by plugin author, which several reviewers flag.
| Integration path | Detail | Source |
|---|---|---|
| API Connector | Call external REST APIs from workflows; the AI Agent can now create API Connector calls | Bubble manual; Sep 2026 AMA |
| Bubble APIs | Data API and Workflow API expose your app to outside systems | Bubble manual |
| Plugins | Marketplace plugins; server-side plugin actions cost 0.2 WU each plus 0.0005 WU per millisecond | Bubble manual (workload activity types) |
| MCP | Editing apps from Claude or ChatGPT is described as coming to beta within weeks of Sep 25, 2026 | Bubble Sep 2026 AMA |
| Payments | Stripe is documented in depth; in-app purchases support subscriptions only | Bubble manual |
For how MCP connectors work in practice today, see our Claude MCP & Connectors guide. For wiring apps together outside Bubble, see our Zapier review.
Workload units: cheap per action, hard to forecast until you have traffic.
Bubble’s manual documents the meter precisely, which is more than many vendors do. A database write costs 0.5 WU, a search starts at 0.3 WU plus 0.015 WU per record and 0.000003 WU per character returned, loading page HTML is 0.15 WU, an external API call is 0.1 WU, and a server-side workflow action is 0.6 WU. Client-side state changes cost nothing unless they hit the server. The catch is composition: one click can run a search, a write, a trigger and an API call.
| Activity (Bubble manual) | Workload units |
|---|---|
| Write or modify one thing in the database | 0.5 |
| Database search (base) | 0.3 |
| Each record returned | 0.015 |
| Load page HTML | 0.15 |
| External API call | 0.1 |
| Server-side workflow action | 0.6 |
| Plan allowances (pricing page) | Free 50K, Starter 175K, Growth 250K, Team 500K per month |
SOC 2 Type II and encryption are documented. Privacy rules are your job.
Bubble’s security page states SOC 2 Type II compliance, a GDPR data processing agreement, TLS in transit, AES-256 at rest through RDS, point-in-time backups, annual pen tests following OWASP WSTG, Cloudflare plus in-house DDoS protection, 2FA for all users and SSO provisioning on Enterprise. Hosting is AWS. Those are the vendor’s statements; I did not see the audit report.
The part no certification can do for you is privacy rules: Bubble data is private only when you write rules for each data type, deciding who can find it in searches, view fields, and create, modify or delete it through the API. The editor includes a Privacy Rules Checker that flags gaps, and the AI Agent runs with your permissions rather than extra ones.
Founders and operators who want a product, not a codebase.
| Your situation | Fit | Likely plan |
|---|---|---|
| Non-technical founder building a SaaS or marketplace MVP | Strong | Starter, then Growth for branches |
| Agency building client apps | Strong, but one plan per client app | Growth or Team |
| Internal tools with logins and workflows | Good | Starter |
| Want iOS/Android plus web | Possible, mobile is beta | Web+Mobile Starter or Growth |
| Developer who wants portable code | Poor | Look at code-generating tools |
| Hobbyist who wants a quick landing page | Overkill | Free |
Seven real, current limitations.
WU is documented but hard to forecast; forum builders have long complained about it and asked for lower rates in Aug 2026.
Each app carries its own plan; one builder noted $32 to build and $134 to get branching, in Jun 2026.
Bubble says data is portable via export and API; I found no app code export in the manual. Moving away means rebuilding.
Capterra reviewers and a third-party review describe a steep curve; months to master, per one review.
Trustpilot is 2.5 of 5 with 66% one-star, citing cancellation and auto-renewal; Bubble replied to 9% of negative reviews.
Native mobile is labeled beta; MCP was announced as coming; BubbleGo is being retired.
I found no published credit or usage model for the agent on the pricing page or in the docs.
Map your reason for looking elsewhere to a tool.
| If you mostly need... | Compare with... |
|---|---|
| A prompt-to-app builder with fast first results | Lovable — Lovable review |
| A browser IDE that hosts and deploys code you own | Replit — Replit review; Bolt.new — Bolt.new review |
| All-in-one AI builder with built-in backend | Base44 — Base44 review |
| UI generation for a React codebase | v0 — v0 review |
| Automation between existing apps rather than a new app | Zapier — Zapier review |
| Front-end only, with a separate backend | WeWeb and Xano are named by forum builders; I did not review them here |
No affiliate relationship shapes this review.
JAVIS has no affiliate arrangement with Bubble. Every CTA points to bubble.io or a live JAVIS review. This is a VERIFIED EDITORIAL REVIEW: I read official docs, pricing and community threads; I did not build an app or run a controlled test.
Build free; pay only when you go live.
Start free on BubbleBubble’s own pricing page.
bubble.io/pricingBubble makes most sense next to the tools that compete with its AI-first rivals.
If you’re deciding between building visually and building by prompt, these are the six reviews I’d read next.
Questions people are actually searching right now
Is Bubble free?
You can build and test for free (50K workload units, development version, API Connector), but you cannot go live or publish to app stores without a paid plan.
How much does Bubble cost?
Web-only plans are Starter $29/month billed annually ($32 monthly), Growth $119 ($134) and Team $349 ($399). Web+Mobile Starter is $59 ($69). Enterprise is custom. Usage beyond the workload included is billed as overage.
What are workload units?
Bubble’s meter of server work: searches, database writes, page loads, workflows, plugin and API calls. A write is 0.5 WU, a search starts at 0.3 WU.
Does Bubble have an AI app builder?
Yes. The Bubble AI Agent launched in beta on Oct 16, 2025, generates and edits pages, data and workflows from prompts inside the editor, and per Bubble’s Sep 2026 recap runs on Claude Sonnet 5.
Can I export my Bubble app’s code?
I found no source-code export in Bubble’s manual. Bubble says data export and open API access keep your data portable. Plan on rebuilding if you leave.
Can Bubble build native mobile apps?
Yes, via a native iOS/Android editor on the same backend, though Bubble’s own blog calls it beta.
Is Bubble secure?
Bubble states SOC 2 Type II compliance, AES-256 at rest, TLS in transit and annual pen tests. App-level security depends on privacy rules you write.
Is there a Bubble MCP server?
Bubble said on Sep 25, 2026 that MCP support was coming to beta within two to three weeks. I could not confirm it had shipped.
Where this came from, and when it was checked.
Pricing and plans: bubble.io/pricing (official, checked Oct 3, 2026) and manual.bubble.io pricing-plans article (monthly/annual table). Overage rate: axonbuild.com (Sep 13, 2026), third party.
Product and AI Agent: manual.bubble.io (introduction, AI Agent, workload articles, version control, privacy rules); bubble.io/blog September 2026, March 2026 and best-AI-app-builder posts; Business Wire and SiliconANGLE for the Oct 16, 2025 launch.
Security: bubble.io/security (SOC 2 Type II, AES-256, pen tests).
Funding and users: TechCrunch, Jul 27, 2021 and Bubble’s own Series A post; user count is a 2021 company claim.
Community: trustpilot.com/review/bubble.io (2.5, 153 reviews, read directly); capterra.co.uk/software/171359/bubble (4.6, 334 reviews, read directly); G2 4.4 from 166 reviews from a search listing and a third-party summary because G2 was not available to us; Bubble Forum threads 397393 (Jun 19, 2026) and 398959 (Aug 2, 2026).
Official video: Bubble channel, video PQNo-VeOu7A (Oct 16, 2025).
Fit score method: JAVIS editorial composite of five 20-point criteria — depth and power 18, AI building speed and maturity 13, price predictability and value 12, portability and lock-in 9, trust, support and security 13 = 65. It is an opinion, not a benchmark.
