Lovable New Features 2026: What Changed (Living Changelog)

A continuously maintained Lovable changelog covering new product features, what changed, and what still has not improved.

This page tracks product updates over time. For the lasting buying and workflow verdict, read the evergreen Lovable review . Quick verdict: what the new Lovable actually changes Lovable has moved from “prompt-to-prototype” into a more serious builder for production web apps. The big shifts are: A unified credit system that pays for build, runtime and AI in one balance. Visual Edits and Supabase Integration 2.0 that cut down routine dev work. Agent Mode and Lovable Agent for multi-step refactors and feature work. File uploads, server-side rendering and usage-based billing for real SaaS. Enterprise governance (Workspace Insights, Security Center, AIUC-1) and distribution into ChatGPT/Claude. It is still opinionated (React/Tailwind/Supabase), web-first and less flexible than IDE-style tools like Cursor or Replit for deep refactors, tests and low-level debugging. Native mobile and fine-grained infra control remain gaps. For teams deciding whether Lovable belongs in an AI stack at all, there is broader context in Best AI Coding Stack for 2026 (By Stage: Prototype, MVP, Production) and a hands-on view in Lovable Review 2026: Fast AI MVP Builder for Operators (But Not Your 3‑Year Stack) . 1. Where Lovable fits in the 2026 AI builder stack Lovable is an AI-driven no-code / low-code app builder. Users describe what they want in natural language and Lovable scaffolds a full-stack web app: React + Tailwind front-end and a Supabase backend by default. Iteration then happens via chat, a visual editor and light dev tools instead of a full IDE, which is why it shows up in most best AI coding tools shortlists as an “AI builder” rather than an IDE. Third-party comparisons consistently frame Lovable as an app builder aimed at non-developers or mixed teams, in contrast to tools like Cursor and Replit which behave as AI-native editors/IDEs. Most of the file tree is hidden; Lovable leans on visual editing and prompts rather than direct code navigation. For teams specifically torn between AI builders and full IDEs, trade-offs are broken down in Lovable vs Replit for Shipping MVPs Fast (2026 Founder’s Guide) and in the broader IDE vs agent stack comparison Cursor vs Claude Code 2026: IDE-First vs Agent-First Coding Stacks . The stack is opinionated. External analyses note that Lovable currently centres on JavaScript/TypeScript, React, Tailwind and Supabase, without arbitrary runtime choice. This is productive if the defaults are acceptable, but limiting for teams that need, for example, Elixir or a custom data layer. Lovable is optimised for: B2B SaaS dashboards and portals. Internal tools and admin panels. Client portals and lightweight CRM-style apps. Marketing/landing sites that benefit from SEO. It does not output native Android/iOS binaries. Independent breakdowns emphasise that Lovable apps are responsive web apps; mobile distribution is via mobile browsers or a PWA/native wrapper built elsewhere. For founders focused on internal tools rather than flagship products, see also AI App Builders for Internal Tools (Not Your Core Product) . If you want an end-to-end playbook from idea to live MVP, there’s a separate walkthrough in From Idea to Weekend Launch: A Practical Lovable MVP Playbook . 2. Pricing and limits: what you actually get 2.1 Unified credit system explained Lovable moved to a unified credit system in 2026: one balance covers building apps, running deployed apps (Lovable Cloud) and AI calls. Instead of separate quotas, workspaces now see a single pool consumed by: Build actions in the editor (generating new screens, refactors, etc.). Runtime infrastructure for deployed apps (Cloud). AI usage both during building and inside apps (model calls). Every plan now includes free monthly credits rather than dollar-denominated Cloud and AI grants. Lovable’s billing update and pricing page describe that all subscribers receive 5 free build credits per day, plus a monthly allocation of free Cloud and AI credits denominated in Lovable credits rather than dollars. Specifically, Free, Pro, and Business workspaces each receive 20 credits per month for Cloud services, and Free workspaces receive 4 credits per month for AI services powering live apps. Above the included free credits, Cloud and AI are billed usage‑based out of the same unified Lovable credit balance that also covers building. Lovable’s billing update explains that the cost of Cloud and AI services is unchanged from the prior system, but the public site does not publish a granular per‑unit rate card for each Cloud or AI operation. Independent pricing breakdowns describe this as a dual‑layer structure: a subscription that unlocks a quota of credits, plus variable Cloud and AI consumption on top. A worked-through cost model with scenarios is available in Lovable Pricing 2026: Plans, Credits, and What You’ll Really Pay . Practically, this simplifies management (one number to watch) but can hide where money is going. Credit consumption is not broken down per action in public docs. Community write-ups suggest that: Generating a new page/screen with AI costs a small number of build credits. Larger multi-file refactors or heavy Agent Mode runs cost noticeably more. In-app AI features draw from the same pool, so popular end-user features can compete with build budget. Because these figures come from community anecdotes rather than official schedules, teams still need to watch usage dashboards and treat credits as a metered cloud resource rather than a flat SaaS feature. 2.2 Plans: Free, Pro and Business Lovable publicly lists three tiers: Free, Pro and Business. External pricing breakdowns add more detail. Plan As of Price & inclusions Free 2026-08-13 $0/month. Lovable’s billing update confirms that all subscribers receive 5 free build credits per day. Free, Pro and Business workspaces also get 20 free credits each month for Cloud services, and Free workspaces get 4 free credits each month for AI services powering live apps. Additional Cloud and AI usage is billed from the workspace’s unified credit balance. Lovable’s pricing page lists the Free plan with unlimited collaborators per workspace sharing a single credit pool. Pro 2026-08-13 $25/month per workspace (USD), according to Lovable’s live pricing grid, with a higher monthly credit allocation than Free and pooled credits across unlimited collaborators on the workspace. The official pricing and billing pages do not currently publish an exact number of included paid credits on Pro or any dollar‑denominated Cloud/AI grant; instead, they describe a unified credit balance plus free daily build credits and a monthly pool of Cloud and AI credits. Business 2026-08-13 $50/month per workspace, per Lovable’s current pricing page. Business includes Pro features plus enterprise‑oriented capabilities such as SSO and Security Center / workspace‑level governance; all collaborators in the workspace share a unified credit pool. Lovable does not publish an exact Business credit allocation or any separate dollar‑denominated Cloud/AI grant in its current official docs. Cloud & AI usage (all plans) 2026-08-13 Under the unified billing system, each workspace receives free Cloud and AI credits expressed in Lovable credits (20 Cloud credits/month on all tiers, plus 4 AI credits/month on Free), and then pays usage‑based rates for additional Cloud infrastructure and AI calls from the same unified credit balance. Lovable’s public docs do not publish a detailed per‑unit rate card for Cloud or AI consumption. All plans allow unlimited collaborators on a project with a shared credit pool. This is friendly for cross-functional teams but also means one heavy user can burn through a shared budget. 2.3 What this means for a small SaaS vs an internal tool Because exact per-credit cost of specific actions is not public, the following patterns are directional rather than precise forecasts: Small SaaS with ~100 beta users : the included free Cloud credits may cover early traffic if the app is relatively light (e.g. CRUD dashboards, occasional AI calls). The free AI allocation on paid plans is limited; if the product itself exposes AI features (chat, summarisation), in-app usage can quickly exceed the free amount and dominate the bill. Heavier B2B internal tool : intensive AI-assisted building (Agent Mode, large refactors) plus modest internal usage can push workspaces into overages even before external launch. Internal teams often accept this as a trade-off for speed, but it requires an internal owner to track usage and cap experiments. Independent write-ups highlight practical constraints that teams encounter: daily build caps on the Free plan, concern about running Agent Mode on large changes, and surprise overages when end-user AI usage spikes. The unified credit view is intended to make budgeting easier than the previous multi-quota system, but it still behaves more like a cloud account than a predictable flat SaaS license. 3. The big product shifts since Lovable 2.0 3.1 Lovable 2.0 as the turning point Lovable 2.0 was a major reset: new brand and UI, multiplayer chat, a more capable dev mode and pricing updates. It marked a shift from “cool demo” to “production apps are expected to run here”. Key changes from that release included: Multiplayer chat : multiple teammates working with the AI on the same project. Dev mode : a more code-aware interface for advanced editing when visual tools are not enough. Updated pricing : setting the stage for the later unified credit model. 3.2 Visual Edits: Figma-like control on generated UI In March 2025, Lovable launched Visual Edits, a Figma-like visual editor that lets users click into generated UIs and tweak text, sizes, colours and layout without re-prompting. For product teams, this changes “day 2 work”: Marketing and ops can adjust copy, spacing and basic layout directly. Designers can tighten a generated UI without dropping into code. Developers are freed from a long tail of minor layout requests. This moves Lovable closer to tools like Webflow on the front-end side while still keeping the AI/code scaffolding underneath. 3.3 Supabase Integration 2.0: less brittle data and auth Supabase Integration 2.0, announced in February 2025, improved how Lovable works with Supabase. The release notes highlight: Automated edge-function log reading. Better user-flow scaffolding around auth and databases. Support for more Supabase capabilities. The practical impact is reduced brittleness around database and authentication flows. Instead of relying on opaque “magic” for data handling, Lovable aligns more closely with Supabase’s own primitives, which helps with debugging and, in the longer run, any migration off Lovable. For teams planning a more serious rollout on Supabase, there is a separate playbook in Lovable + Supabase Production Setup: Safe Integration for Real Users . 4. Agent Mode and AIUC-1: from vibe coding to governed agents 4.1 Agent Mode and “Lovable Agent” Agent Mode (beta) arrived in July 2025 as an autonomous AI that can think, plan and take actions across a project instead of only responding to one-shot prompts. Rather than asking for a single screen, teams can request multi-step changes such as: “Add usage-based billing to this app.” “Refactor the auth flow to support organisations instead of just users.” “Migrate the dashboard to use the new schema.” Agent Mode can decompose the request, plan sub-tasks and execute across files, similar in spirit to agentic coding assistants in IDEs but constrained to Lovable’s environment. In July 2025, Lovable announced “Lovable Agent” alongside a $100M ARR milestone, positioning the platform more explicitly around agentic coding rather than just prompt-to-app scaffolding. Public material suggests Lovable Agent builds on Agent Mode capabilities for more autonomous multi-step work. 4.2 AIUC-1 certification In July 2026, Lovable became the first AI coding agent platform to earn AIUC-1 certification, a security, safety and reliability standard for AI agents. For founders selling into regulated or security-sensitive customers, this matters for two reasons: It signals that Lovable’s agentic features (such as Agent Mode) are subject to formalised safety and governance controls. It gives procurement and security teams an external reference point when assessing the risk of letting an AI modify production code and infrastructure. AIUC-1 does not remove the need for a separate security review, but it changes the baseline conversation from “experimental AI toy” to “agent platform with audited controls”. 4.3 Enterprise governance: Workspace Insights and more Workspace Insights, launched June 2026, provides a Security Center view for all projects in a workspace, with visibility and governance features for admins. The July 2026 enterprise roundup lists additional features aimed at larger teams: reusable authentication, publishing controls, security scanning and cleanup of abandoned apps. Together, these additions shift Lovable from a solo-builder tool toward something a mid-size team can standardise on: Admins can see every app, who owns it, and its security posture. Publishing controls curb “shadow apps” going live without review. Security scanning and cleanup reduce the tail of forgotten prototypes. Combined with AIUC-1, this is Lovable’s clearest move into enterprise and regulated use, though teams still need to check data residency, model governance and internal compliance requirements. 5. Lovable apps inside ChatGPT and Claude In July 2026, Lovable announced that apps built on the platform can be used inside ChatGPT and Claude. Instead of exposing a product only as a standalone web app, teams can make it available as a tool within major AI assistants. This unlocks new patterns: End-users interact with functionality from within their preferred AI assistant, never visiting a separate URL directly. The Lovable backend handles business logic, data and billing, while ChatGPT/Claude serve as the interface. For internal tools, employees can stay in their AI workspace and call the app as-needed. However, not everything moves into the assistant: Teams still define data models, workflows and UI flows in Lovable. Debugging is more complex: implementers are now reasoning about the assistant’s behaviour plus the app logic. Rate limits and usage policies of the host assistant apply, which may constrain some high-frequency or high-volume use cases. For founders, the main implication is distribution. It becomes possible to ship a Lovable-built service that “lives” where users already spend time, with Lovable handling much of the underlying stack. If you are choosing between assistants, ChatGPT Free vs Plus vs Pro in 2026 and Claude AI Review 2026 cover how these ecosystems differ for app distribution. 6. Models and “brains”: GPT-5, Gemini 3.6 Flash and multi-model apps 6.1 Model menu and defaults Lovable supports multiple AI model families for both build-time and in-app features. A comparative technical report lists Lovable as supporting OpenAI GPT-4.x/5, Anthropic Claude, Google Gemini, Llama models via Groq and Cohere Command, among others. This aligns with Lovable’s positioning as a model-agnostic platform where teams can choose the “brain” for each task. From July 22, 2026, changelog summaries note that Lovable’s default “AI brain” for built-in tools is Gemini 3.6 Flash when no specific model is selected. This suggests a preference for fast, low-cost generation as the baseline. 6.2 GPT-5 integration Lovable announced GPT-5 availability via a limited preview integration on August 14, 2025. Builders can access GPT-5 both: While building apps (e.g. more capable code generation and refactors). As an in-app AI feature exposed to end-users. This is relevant for teams that want to offer frontier-model capabilities (reasoning, long-context workflows) inside a no-code-built product without wiring OpenAI’s APIs directly. 6.3 Per-task model selection and trade-offs Lovable’s documentation and external analyses indicate that teams can choose models per task rather than only globally. For example: Use a fast, cheaper model (e.g. Gemini 3.6 Flash) for UI text tweaks or low-stakes recommendations. Reserve GPT-5 or Claude for complex reasoning, summarisation or multi-step planning. The main trade-offs mirror direct API usage: Latency vs cost : faster, cheaper models for interactive flows; heavier models for batch or high-value decisions. Hallucinations and reliability : more capable models may still require guardrails and validation in app logic. Budgeting : because all AI usage draws from the same credits, misconfigured in-app prompts can silently burn through build capacity. Lovable abstracts away client libraries and model-specific plumbing but does not remove the need for experimentation around prompts, error handling and guardrails. 7. New features that change what you can actually ship 7.1 File uploads and app-from-data workflows On September 29, 2025, Lovable launched File Uploads, allowing users to drop files directly into Lovable and generate apps or websites from them. This broadens input beyond text prompts: Lovable’s Visual Edits feature, shown here on the official announcement page, backs up the claim that teams can tweak generated UIs directly on the page instead of constantly re-prompting or dropping into code. Upload a CSV of transactions and generate an internal analytics dashboard. Upload a folder of documents and generate a searchable portal. Upload content assets and generate a marketing site. For non-technical operators, this reduces the gap between having a dataset and having a usable tool around it. 7.2 Server-side rendering and SEO Community updates report that Lovable apps are now server-side rendered and indexable by search engines and AI answer engines, a change from earlier purely client-side builds. This matters if a Lovable app is outward-facing: Marketing and content pages can rank in search, not just serve logged-in users. AI answer engines can surface content more reliably. Lovable becomes more plausible for SEO-sensitive microsites and content apps. 7.3 In-app usage-based billing Lovable has shipped tooling for in-app usage-based billing (usage-metered plans, real-time wallets, billing portals) that can be wired into a Lovable app mostly via prompts, according to community posts. As of the sources cited above, Lovable’s core documentation and public blogs do not provide a detailed, centralised specification of every billing workflow pattern, so implementation details rely partly on community experimentation.