Opportunities › Cloud infrastructure for AI agents › Proposed app
Runpen
Isolated VM per agent run, billed only for the seconds the agent works
- Score
- 54
- confidence 60
- Evidence
- 9
- items on the page
- Rivals
- 1
- named in sources
- First version
- 60 d
- estimate
- Type
- Desktop and mobile
- PWA: on computers and phones alike
- Market since
- 2024
- New market · AI estimate
ScreensConcept
Desktop and mobile: Engineers set up spend limits and policies on desktop, but also use the phone-ready kill switch to stop runs on the go, making both platforms essential. Sample screens on a computer and a phone, in the proposed brand. Their numbers are sample data.
Runs
All isolated3 running
About this appAI proposal
Runpen gives every AI agent run its own fully isolated virtual machine. The machine starts when the agent has work, stops when it is done, and you are billed for the compute time in between. No always-on servers, no shared kernel between runs.
It is for teams running agents in production who are stuck choosing between general-purpose cloud VMs and serverless. According to the opportunity material, those either charge for idle time or offer sandboxing that is too weak or too heavy.
A PWA dashboard shows each run, its cost and its network activity, so an operator can check or kill a run from a phone. Runs start from an API call; the dashboard is where you set limits and watch spend.
- Tagline
- A sealed room for every agent run. Pay only while it works.
- Voice
- PlainPreciseCalm
- Why this name
- A pen is a small enclosed space for something that needs containing. Run is what an agent does. Two plain words: one sealed space per run.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
One VM per run
Each agent run gets its own machine with its own disk and network. When the run ends the machine is destroyed, so nothing carries over to the next run.
No member yet
Pay for active seconds
Billing counts compute time while the agent runs. Idle gaps between bursts cost nothing, because no machine is kept running.
No member yet
Run timeline
See every command, file change and network call of a run in order, so you can find out why an agent did what it did.
No member yet
Spend and time limits
Set a maximum duration and cost per run, per agent and per day. Runs that pass a limit are stopped and flagged.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#1F6F5CPine · primary#F6F1E7Paper · surface#1F2A24Ink · text#E2762DEmber · accent#3A5BA0Slate blue · data
Typography
Aa
Runpen
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
A sealed room for every agent run. Pay only while it works.
PricingAI proposal
The material gives no competitor prices, only that general-purpose clouds charge for always-on compute. These tiers are a starting point to test; the per-second rate must be set from real costs.
Try
Free
- Small monthly compute allowance
- Up to 3 agents
- 7-day run history
Team
$49 a month
- Larger compute allowance, then per-second billing
- Unlimited agents
- Spend limits and alerts
- 30-day run history
Scale
Contact us
- Volume pricing on compute time
- Network allowlists and audit export
- Higher concurrent run limits
- Support with response times
Would you pay Team · $49 a month?
Your answer tells PWAS which prices people accept. It changes no score.
No answers yet.
What it could be worthEstimate
What a software business making this much sells for today, from real deals. Pick a monthly revenue.
If it makes
Worth about
$194,000
Sale range $132,000 to $263,000
$120,000 a year in revenue
From 87 software businesses listed and sold on Empire Flippers in the last year, revenue vetted by the broker: they sell for about 19× their monthly revenue, most between 13× and 26×.
Measured on profit, the middle deal is 32× monthly net profit: about 2.7 years of profit.
An estimate, not an offer or financial advice. A real price depends on profit, growth, churn, how much of the work is the owner's, and the buyer.
How it would growAI proposal
How the product would reach new customers through its own use, and how the first ones would be found. A plan to test, not a result.
- Built-in loop
- No loop in the product. Runpen runs behind the customer's own product, and their end users never see it. It reaches new customers through developer content, docs, open examples and direct sales to teams already running agents.
- Do they become customers?
- No built-in loop. To test: whether engineers who find Runpen through a how-to guide or benchmark write-up start a free account and move to a paid Team plan within 30 days.
- First customers
- Engineering leads at startups of 5 to 50 people that already run AI agents in production, such as support triage or browser automation, on rented VMs or serverless.
- Where to reach them
- Hacker News and dev forums: Post a technical write-up on isolation and per-second cost, with a public cost comparison for a sample bursty workload.
- Agent framework communities: Publish ready-made Runpen runners for popular agent frameworks and answer questions in their chats.
- Direct outreach: Message teams that posted about agent sandboxing or cloud bills and offer a free week of migration help.
- First milestone
- Five teams paying for Team or usage within 60 days, each running agent workloads in production.
Why people already pay for this
- Public claimFounders have paying customers and approximately 700 people on the waitlist, indicating market demand.
- Public claimThe product successfully demonstrated agent capability on a standard benchmark, completing 82% of 369 automation tasks.
- Public claimExisting cloud providers lack adequate sandboxing for agents or charge for idle time during bursty workloads.
- Public claimPurpose-built infrastructure with pay-per-use pricing solves both isolation and cost efficiency for agent workloads.
- Public claim
Tech stack
Every proposal builds on the same PWA base, installable from the browser on any phone or computer. Dashed boxes are what this app adds.
- AppNext.js and TypeScript
- PWAManifest and service worker
- UITailwind CSS
- Data and authSupabase Postgres
- PaymentsStripe
- EmailResend
- HostingVercel
- This app · Isolation runtimeMicroVM hypervisor on bare-metal hostsFull VM isolation per run with fast start needs lightweight virtual machines, which Vercel and Supabase cannot host.
- This app · Compute hostingBare-metal or nested-virtualization cloud capacityThe product sells compute time, so it needs machines it can start and stop within seconds.
RisksAI inference
- TechnicalFast VM start, clean teardown and escape-proof isolation are hard to get right. One breakout would end trust in a product that sells security.
- CommercialMargins depend on buying compute and reselling it by the second. Large clouds could add similar per-run VMs and undercut a small team.
- CommercialDemand rests on public claims from one forum thread, including a waitlist of about 700. Whether teams pay a new vendor instead of building their own is untested.
- PlatformUpstream hosts and model providers may restrict or reprice the capacity this depends on, which would change unit costs.
- RegulatoryAgents may handle customer data inside Runpen machines. Buyers will ask for data-handling terms and audit evidence before they put real workloads on it.
Everything marked AI proposal or AI inference was written by a model from the evidence on the opportunity page. The proposed name must be checked against trademark registers before it is used.