Artificial Intelligence

Lovable Ditched the 30-Minute Agent Limit

Lovable split agent memory from compute to survive 700 daily deploys and run agents for 10 hours.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

2 min read

Lovable co-founder and CTO Fabian Hedin told AI Engineer that Lovable deploys 700 times per day, about once every two minutes, and that forced an odd design choice.

Lovable Ditched the 30-Minute Agent Limit - AI Engineer
Lovable Ditched the 30-Minute Agent Limit, AI Engineer

Hedin said Lovable split the agent process from agent memory so a 10-hour run can survive a deploy. The team once capped agents at 30 minutes because longer runs were unreliable. Each agent now gets an inbox for human, agent and scheduled messages. The agent copies those into an append-only trajectory modeled on Git, Hedin said, where events point to a parent and can fork without copying shared history.

The speed trick is checkpointing the computer, not just the chat.

Hedin put the start at a golden checkpoint with OS, system services and a ready dev server, copied in about 230 milliseconds, with a fresh checkpoint saved after each agent turn for subsecond restore. Forking that machine powers drafts and parallel tries, Hedin said, with copy-on-write on RAM plus streaming boot so the OS runs before all memory arrives from cold storage. Lovable also rewrote its dev server in Rust, Hedin said, cutting preview startup from 40 seconds to three seconds with code starting two times faster in 6.5 times less memory.

Verification runs in two loops, Hedin said, with an inner loop during the turn for build, lint, type and runtime errors plus a headless browser check. The outer loop between turns uses production behavior, A/B tests and business metrics like payments and conversions. The agent has browser and GitHub access and plugs into hosting, auth, databases, payments, email, SEO and ads, Hedin said, with proactive business operation tools planned for 2027 as reported by Runtime Wire. Hedin put current load at over a billion tokens per minute.

You've felt the failure mode if you have used chat-to-app builders. The preview looks fine while the bug lives in auth or data and the agent keeps patching. Lovable is betting long runs work when the machine restores in milliseconds and history forks without a restart, which leaves you managing outcomes and rollbacks instead of files.

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsLovableFabian HedinAI EngineerGitGitHubRustGoogleMeta

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.