Skip to content

Build1 publisher3 min readPublished

A five-minute function ceiling pushes long agent loops off Vercel's Hobby tier

Fluid compute caps Hobby functions at five minutes and Pro at 800 seconds, so an agent working a long task list needs a queue or a different host. The cheap VPS escape hatches have repriced too.

The Engineer · Build desk

Illustration accompanying A five-minute function ceiling pushes long agent loops off Vercel's Hobby tier

What happened

  • The dev.to hosting guide says a $20 Vercel bill becomes a $200 bill the first month an agent loop runs hot, on a model that charges for invocations, GB-hours and bandwidth.
  • Hetzner raised its cloud prices on 15 June 2026, and the shared CX plans the guide originally recommended are no longer in its price list.
  • A netcup VPS 500 G12 with 2 vCores and 4 GB RAM lists at EUR 5.91 a month, which the guide says is enough to run production Mastra agents.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • decision The duration ceiling settles the code's shape before the framework does: anything that runs past the limit gets a queue and a poll loop, and that is a rewrite, not a config change.
  • cost Comparing provider list prices directly overstates the cheap option, because Hetzner's number excludes VAT and the IPv4 address while netcup's includes both.
  • exposure A self-hosting cost model built on a bookmarked plan is exposed to the provider editing its price list, which is what happened to the CX tier in June.

Fluid compute is on by default, and with it a Hobby function stops at five minutes [1]. Pro raises that to 800 seconds by configuration, and to 30 minutes on the extended-duration beta [2]. Measured against Hobby's 300 seconds, those are 2.7 times and 6 times the headroom [3]. The limit is a hard maximum on the function, so an agent still working its task list at that point needs background jobs or a queue-and-poll pattern, and the dev.to guide's position is that those workarounds erase the simplicity argument [4].

That is where framework choice and hosting choice stop being separable. The AI SDK treats Next.js as the primary host, with server actions and route handlers as the recommended patterns, and the guide says you start fighting the framework the moment the backend lives elsewhere [5]. Mastra's deploy command targets Cloudflare, Vercel, Netlify or self-hosted Node [6]. LangChain is Python-first with a TypeScript port that is always one version behind [7], so its hosting question is LangGraph Cloud or a Python-capable box you run yourself [8].

Vercel bills on function invocations, GB-hours and bandwidth, and the guide's reading is that streaming consumes the bandwidth while loops consume the invocations [9]. For the jump from $20 to $200 to show up in your own account, your traffic has to resemble the example the guide names: a chatbot with a thousand daily users [10][11]. A single-turn completion endpoint at the same request count moves all three meters differently. The guide does not publish the invocation counts or GB-hours behind the figure.

The VPS side needs its own correction before the prices compare. netcup lists EUR 5.91 a month for the VPS 500 G12, 2 vCores and 4 GB RAM, including 19% German VAT and with IPv4 included [12][13]. Hetzner lists excluding VAT, with the primary IPv4 address billed separately at EUR 0.50 [14]. Strip the VAT out of 5.91 and you get 4.97; take off the address charge and a Hetzner listing has to come in at or under roughly 4.47 before VAT to beat netcup on a like-for-like line [15]. The shared CX plans this guide originally recommended are no longer in Hetzner's price list following the 15 June 2026 increase [16].

The other starting point the guide names is a DigitalOcean Basic droplet with 2 GB at $12 a month [17], which is half the memory of the netcup box [18]. Every VPS price in the guide was re-checked on the providers' own pages on 21 September 2026 [19], which is more diligence than most hosting roundups carry a date for. For Mastra specifically the guide recommends Cloudflare Workers when agents are short-lived and a small VPS when they run long or need persistent state [20].

What to watch

  • Whether Vercel's extended-duration beta reaches general availability and lifts the 800-second Pro configuration ceiling.
  • Whether Hetzner's next price list edit restores shared plans in the range the guide used to recommend.
  • Whether Mastra adds a first-party host for long-running agents, removing the small-VPS step from its deploy advice.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories