Quick answer
The local whoburnedmore CLI command is npx whoburnedmore --local. Decline its optional publish offer to keep the report local. The standard command signs in and submits daily aggregates. Neither mode submits prompts, source code, or filenames.
What this guide answers
The reader wants to understand Lovable AI Gateway cost and deployed app usage.
Best for
Lovable builders adding AI features to apps.
Key facts, checked November 8, 2026
- Lovable AI is usage-based and priced per model on tokens, images or video seconds. Free, Pro and Business include a small monthly AI grant (4 AI credits), which Lovable calls temporary. source
- No single model is the default. Lovable suggests a starting model per job: GPT-6 Astra for chat, Gemini 3.1 Flash Lite or GPT-6 Luna for high-volume text, and Gemini Embedding 2 for embeddings. source
- A 429 'Too Many Requests' error means a per-workspace rate limit was hit. A 402 'Payment Required' error means the workspace is out of credits. Canceled requests may still be billed. source
- The per-project AI usage dashboard keeps 24 hours of history on Free and 90 days on paid plans. source
Separate build time and runtime
Build credits help create the app. Gateway usage happens when the app's AI features call models after deployment.
- Builder prompts
- App model calls
- User-triggered usage
Meter every AI feature
Any button that calls a model needs a usage story. Without limits, one popular workflow can consume the budget quickly.
- Per-user caps
- Daily caps
- Admin alerts
Watch prompt size
Runtime cost is shaped by prompt and response tokens. Keep prompts small and avoid sending full records when summaries will do.
- Trim context.
- Cache repeated answers.
- Use cheaper models where acceptable.
Add AI cost to product analytics
Production AI usage should be reviewed with product metrics. The right question is whether a feature's cost is justified by user value.
- Cost per active user
- Cost per feature run
- Conversion impact
Handle 402 and 429 in the app, not just the budget
When the workspace runs out of credits, the AI calls in your deployed app start returning 402. When traffic bursts past the per-workspace rate limit, they return 429. Lovable doesn't publish the rate-limit numbers. Both errors reach real users, so every AI feature needs a fallback path. Show a clear message, queue or retry 429s with backoff, and stop retrying on 402 entirely, because a retry loop on 402 only produces more failed calls.
- 429: back off and retry, then degrade gracefully
- 402: stop calling, alert the owner, show a non-AI fallback
- Remember that canceled requests may still be billed
Pick the model per job, then cap it
Because pricing is per model, the biggest lever is choosing the smallest model that passes for each feature. Classification, tagging and short summaries rarely need the chat-default model. Lovable's own guidance points high-volume text work at lighter models. Then use paid-plan usage limits, settable per project, member, group or workspace, to block or alert. A production project with public AI features should have a blocking limit, not just an alert.
- High-volume text: a lighter model
- User-facing chat: the chat model, with a per-user cap in your own code
- Embeddings: the embedding model, computed once and stored
Separate runtime spend from development spend
Run credits are product cost and scale with your users. Build credits, and any local agent tokens spent on the repo, are development cost. whoburnedmore covers only the development side, and only for local agents such as Claude Code, Codex and Cursor. It does not read Lovable credits or app runtime usage. Review the two in separate meetings: runtime cost with product metrics, development cost with the engineering budget.
Common questions
How is Lovable AI billed?
By usage, priced per model on tokens, images or video seconds, from the workspace's single credit balance. Lovable shows it as Run credits in Usage details.
What does a 402 error from Lovable AI mean?
The workspace is out of credits. Top up or raise the limit, and make sure your app stops retrying and shows a fallback.
Which model does Lovable AI use by default?
There is no single default anymore. Lovable suggests starting models per job, for example GPT-6 Astra for chat and lighter Gemini or GPT models for high-volume text.