https://lovable.dev/
Our editorial read
A practical UI-first builder for validating product flows before engineering the full system.
Best fit
Founders and product teams that can describe screens, roles and workflows clearly and want a fast visual iteration loop.
Where we would hesitate
Do not confuse a polished front end with finished permissions, data validation or operational readiness.
Decision rule
Choose Lovable when user-flow clarity is the constraint; choose a code-first builder when backend behavior and extensibility dominate.
Evidence note: This judgment synthesises the linked official documentation, release notes and product context. It is not a claim that AI Tools Manual independently benchmarked every tool.
Recent official changes
Lovable’s Aug 20 changelog groups several product changes: Gemini Embedding 2 becomes the default for new semantic search and RAG builds; design systems reach all paid plans; publishing becomes a clearer two-step flow; and connected service accounts can be used privately for project-scoped inspection and testing. Verify plan scope, connector permissions, approval prompts, and any billing or security implication before publication. We did not independently test the features.
Primary source
August 2026: cost visibility, connectors and audit controls
Lovable’s August changelog adds several practical controls. Users can see time and credits spent while a message is still running; Enterprise audit logs can be filtered by client; previews retry when a connection drops; and LinkedIn skills are limited to Free and Pro workspaces. The changelog also adds per-user GitHub access for published apps, so each app user can connect their own account instead of sharing one builder credential. Other entries add the Chat Latest model alias, MCP registry setup, Figma import, and design systems on paid plans.
These changes improve visibility and isolation, but they also create configuration decisions. Confirm the workspace plan, whether a feature consumes credits, the OAuth scopes and repository boundaries for per-user GitHub access, and what audit events are retained. Treat openai/chat-latest as a moving alias when predictable cost or output matters; pin a numbered model where possible. This summary follows Lovable’s official changelog and does not claim an independent test of deployment, billing, or connector security.
Primary source: Lovable changelog.
Recent official changes
Lovable’s Aug 20 changelog groups several product changes: Gemini Embedding 2 becomes the default for new semantic search and RAG builds; design systems reach all paid plans; publishing becomes a clearer two-step flow; and connected service accounts can be used privately for project-scoped inspection and testing. Verify plan scope, connector permissions, approval prompts, and any billing or security implication before publication. We did not independently test the features.
Primary source
September 2026 update: domains, connectors, and longer-running builds
Lovable’s official changelog lists several changes that matter to teams building and publishing apps. From September 1–2, Lovable added in-chat custom-domain setup, a mobile-app download area in Settings, Enterprise support for Lovable in Slack, and connectors created from a pasted API key. Domain purchase is limited to workspace admins and owners on paid plans, and the product shows a confirmation card before buying or connecting a domain.
The changelog also says workspace admins and owners can create custom REST API connectors on any plan, with credentials and optional knowledge files; members can then reuse the connector across projects. Priority processing now works with supported Gemini chat models at a premium price. Lovable can work on a single build message for up to 10 hours, while credit check-ins can still pause the task. These details are provider-reported and may depend on workspace, plan, region, or rollout state.
What to verify before adopting the new workflow
Buying or connecting a domain changes ownership and billing, so verify the workspace role, checkout account, renewal terms, and DNS handoff before confirming. A connector created from a pasted key may store that key as a project secret even when a reusable workspace connection is created; review who can use it and rotate the key if it was exposed in chat. For GitLab, Slack, Firebase, and other connectors, check whether access is shared at workspace level or per user. Longer-running builds can consume more credits than a short message, so set a budget and inspect the resulting diff.
Our editorial judgment
We see the update as useful for teams that already have a clear workspace owner, a secrets policy, and a staging deployment. The domain flow reduces setup friction, and custom connectors can avoid rebuilding the same integration for every project. The convenience is not a substitute for least-privilege permissions or a code review. For a solo prototype, the safest path is to use a non-production domain, a narrowly scoped key, and a disposable test project.
Best fit: paid workspaces with an administrator, a staging environment, and a documented process for secrets, domains, and connector access.
Not fit: teams that cannot review API scopes, users who expect free unlimited long-running builds, or anyone planning to paste production credentials into a chat without rotation and audit.
Decision checklist
- Confirm the workspace role and plan before buying a domain or enabling a premium connector.
- Use a test domain and least-privilege API key; record who can reuse the connection.
- Review the app diff, preview, and connector logs before publishing.
- Set a credit ceiling for long goals and require a human checkpoint before production deploy.
- Document ownership, renewal, DNS, and offboarding steps.
See our Lovable tool profile, AI builder security guide, and generated-code review guide. Next review: October 3, 2026, or sooner if Lovable changes domain purchasing, secret storage, connector permissions, credit pricing, or plan access.
Evidence note: this update uses Lovable’s first-party changelog and editorial risk analysis. We did not perform an independent build, connector, billing, or security test.
September 2026 official update: Sutro and Salesforce
Lovable’s official blog lists a September 18 announcement that it acquired Sutro, with the Sutro founder and three engineers joining Lovable. On September 15, Lovable also announced Salesforce support so apps and agents can run through Headless 360, Slack, and installable agents. A separate September 15 post says Lovable is gradually moving previews to its open-source OJ Rust-native development server for faster startup and lower memory use.
Who should care
Teams building internal tools may see a clearer path from a prompt-built prototype to a connected business workflow, but the announcements do not mean every Salesforce connector or OJ-powered preview is available to every account. Enterprise buyers should ask which permissions are required, where data is processed, how OAuth scopes are reviewed, and which features remain in rollout.
Editorial decision
We would shortlist Lovable for product teams that want rapid full-stack prototypes and can involve an engineer in connector and deployment review. We would not recommend treating the acquisition as proof of production reliability; roadmap integration and security evidence still need verification. Link this review to the practical guides before choosing a builder, and use the comparison hub to compare Lovable with Replit and v0.
Primary sources: Lovable blog and Lovable legal change log.
October 1, 2026: live voice, EU inference and API-key allowlists
Lovable’s changelog records three changes with different implications. First, GPT Live can power live voice conversations inside apps built with TanStack Start; the changelog says users can interrupt the conversation and have the app act on what they say, and that the feature uses credits. Second, Enterprise workspace admins can enable EU inference so workspace model requests are handled by providers in the European Union, including requests made while building and AI features in published apps. Lovable warns that some models or features are unavailable under this setting and that it consumes more credits. Third, Business and Enterprise admins can restrict workspace API keys to approved IP addresses and ranges.
Why this matters: GPT Live makes voice interaction a more practical application pattern, but the model and workflow should be tested for latency, interruptions, and failure handling before it is used in a customer-facing flow. EU inference is a data-residency control with trade-offs, not a blanket compliance certification or a guarantee that every data path is regional. IP allowlisting can narrow where workspace API credentials are accepted, but it does not replace key rotation, least privilege, or application-level access controls.
Who should evaluate these changes—and what needs human verification
App builders with an established voice use case can prototype GPT Live on a low-risk workflow and review credit consumption before expanding. EU-regulated or region-sensitive teams should ask their privacy and security owners to confirm which model providers, features, logs, and published-app requests remain eligible when EU inference is enabled. Workspace administrators should inventory integrations that use Lovable API keys, then test the allowlist from approved and unapproved network locations before enforcing it broadly.
Do not assume the three announcements answer the same governance question. The changelog does not enumerate all excluded models, define every retained-data path, or establish that EU inference alone satisfies a legal obligation. It also does not explain how key restrictions interact with proxies, rotating egress addresses, or every deployment environment. Check the current EU inference documentation, API key documentation, and AI features and model documentation before making a procurement or compliance decision.
Our assessment and evidence disclosure: Taken together, the updates broaden Lovable from a prompt-to-app builder toward configurable AI features inside the apps it creates, while adding enterprise controls that still require policy review. They are useful capabilities, not proof of secure or compliant deployment by default. Primary sources: Lovable changelog, October 1, 2026 and the linked official documentation. AI Tools Manual did not test the voice feature, inspect a tenant’s data path, validate regulatory compliance, or conduct penetration testing. Review our AI privacy checklist and Replit vs. Lovable comparison. Recheck model eligibility, region behavior, credit use, and key enforcement before approval.
September 30, 2026: Lovable reports a TanStack Start security advisory
Lovable says its security team identified CVE-2026-102989 in TanStack Start, a framework used by some Lovable apps. The company describes a crafted-link attack that could cause an affected app to run unwanted JavaScript in a visitor’s browser. It says exploitation required both affected server functionality and a visitor opening the malicious link; using TanStack Start alone did not make every project vulnerable. Lovable also says it found no evidence of exploitation in the logs it reviewed. That statement is limited to the logs and period the company examined, not an independent audit of every deployment.
For hosted Lovable apps, the company says it deployed firewall rules while the upstream fix was prepared, and that the next edit automatically upgrades an affected project. It also says owners can apply the fix earlier at no credit cost through the project Security page, and that new projects use the patched version. Lovable’s own hosting protections do not cover an app deployed elsewhere; self-hosted owners need to apply the upstream update and redeploy through their own hosting setup.
Action checklist for Lovable users
Best fit: this update matters to anyone with a Lovable project, especially projects using server-side routes or deployed outside Lovable hosting. It is not evidence that every app was compromised. Our assessment is to check Lovable’s affected-project notice and Security page, apply the available update, and confirm the deployed version. Self-hosted teams should identify their TanStack Start dependency, follow the upstream advisory, redeploy, and review logs for suspicious activity appropriate to their application.
Do not rely on the hosted firewall if you export or self-host the project, and do not interpret “no evidence found” as proof that no exploitation occurred. For broader app-builder context, compare our Bolt.new review. Evidence note: this summary follows Lovable’s official security notice and upstream advisory; it is not a security audit or legal advice. Next review: October 15, 2026.
October 2026 workspace controls: domains, cross-chat tasks, and secrets
Lovable’s October changelog adds several workspace-level capabilities with different implications. On October 9, Enterprise workspace admins and owners gained a publishing-domain setting for project subdomains; existing lovable.app addresses continue to work. Lovable also added a way to reference another chat so it can read that discussion or receive a task, and a generated-secret flow for values that must be entered in two places. The company says a generated secret is shown only in the form and is not repeated in project chat, logs, or code.
On October 8, Free and Pro users gained invite-only publishing for apps, with invited visitors required to log in. Workspace publishing to the whole workspace or groups still requires Business or Enterprise. On October 7, paid-plan workspace admins and owners gained configurable credit thresholds, alerts, and blocking actions that can pause building, Lovable Cloud, AI features, or metered connector requests.
Who should care: teams publishing private previews, handling shared credentials, delegating work across conversations, or managing shared credits should review these controls before changing their workflow. Invite-only access is not a substitute for testing authorization inside an app. Cross-chat context sharing can expose prior project discussion to another task, so users should avoid referencing chats containing information that should not flow into the new context.
Editorial assessment: the updates make Lovable more useful for governed team projects, but they also make workspace settings and permissions more consequential. The secret-handling and credit-blocking statements above are vendor descriptions; AI Tools Manual has not independently tested them. This update affects privacy, security, access, and spend controls, so it is held for human review.
Human review required: verify plan eligibility, login/access behavior, cross-chat data handling, secret visibility/retention, and which actions a credit block interrupts. Primary source: Lovable official changelog (October 7–9, 2026).
October 15, 2026: Enterprise audit exports and managed Claude access
Lovable’s October 15 changelog adds scheduled audit-log exports for Enterprise workspaces to S3, Google Cloud Storage, Azure Blob Storage, Splunk or Azure Log Analytics. The documentation says the export can extend retention beyond Lovable’s 13-week log window, with new events delivered hourly or every three hours. Lovable states that it does not charge extra for this export; destination storage and ingestion are billed by the provider. The same date adds enterprise-managed authentication for using Lovable’s MCP server in Claude through a company’s SSO. See the official changelog, export documentation and authentication prerequisites.
This matters to security and IT teams that need an audit trail in their existing SIEM or want identity policy to govern Claude-to-Lovable MCP access. It is not a general consumer feature: exports are described for Enterprise, while managed authentication requires Business or Enterprise, Okta configuration, enabled third-party MCP access and compatible Claude organization settings. The docs also warn that an existing SSO application should not be replaced just to expose Cross App Access controls.
Before deployment, test an export destination, verify event completeness and retention, confirm administrators and members have the intended permissions, and check the separate Claude/Okta subscriptions and Lovable credit consumption. We have not implemented or security-tested the flow. Compare this UI-first builder with Replit or v0. Next review: November 16, 2026.
Human review required: validate plan eligibility, 13-week retention, event coverage, export cadence, destination-provider charges, Okta/Cross App Access prerequisites, Claude support, account linking and credit/subscription boundaries.nn
What changed in Lovable on October 20
Lovable’s October 22 changelog says Gemini 2.5 Pro, Gemini 2.5 Flash, and Gemini 2.5 Flash Lite were retired for apps’ AI features on October 20. The changelog says requests for Pro and Flash now route to Gemini 3.8 Flash, while Flash Lite requests route to Gemini 3.1 Flash Lite. Lovable says existing apps continue to work without a code change, but the replacement may cost a different amount per request and produce different responses.
This is a provider-side model substitution, not merely a new option in the editor. An app can remain technically available while changing its output, latency, or usage cost. Teams using one of these model IDs should inspect the current model configuration and test important prompts against the routed replacement before relying on unchanged behavior.
Who should act and what to test
App owners using Lovable AI features with the retired Gemini 2.5 IDs should inventory affected workflows, capture representative prompts and expected outputs, and check usage costs after the transition. If an output difference matters, ask Lovable to select a current model deliberately and rerun the evaluation. For customer-facing or high-impact features, add a human review or fallback path instead of assuming the replacement is equivalent.
Lovable states the new routing, but it does not establish that the replacement models are functionally identical for every app. This update is therefore held for review because a model change can alter both user-facing behavior and billing. Related reading: Bolt.new’s model and platform update. Next review: November 11, 2026.
Evidence note: Dates, model IDs, and routing descriptions are taken from Lovable’s official changelog; no independent test was run.