Opportunities › Persistent cloud compute for autonomous agents › Proposed app
Perchline
Persistent, isolated cloud machines your agents can launch, track and cap themselves
- Score
- 56
- confidence 61
- Evidence
- 13
- items on the page
- Rivals
- 3
- named in sources
- First version
- 60 d
- estimate
- Type
- Mobile app
- PWA: installs on iPhone and Android
- Market since
- 2006
- Mature market · AI estimate
ScreensConcept
Mobile app: Engineers check fleet status, approve agent requests, and view cost summaries in a phone-friendly console while away from their desk managing background workloads. Sample screens on a phone, in the proposed brand. Their numbers are sample data.
Fleet
42 running
- coder-pr-1182 vCPU · 4 GB
- rl-env-sweep-07H100 · 3d 4h
- scraper-nightlySuspended
- eval-batch-228 vCPU · 32 GB
- Spend today$63.40
About this appAI proposal
Perchline gives each AI agent its own persistent cloud VM, so long coding sessions, training runs and RL environments keep going when your laptop sleeps. You or an agent start a machine from a CLI, and a phone-friendly console shows the whole fleet, what it costs and what each machine can reach.
Built for teams that run several agents at once, ML engineers with multi-day jobs, and companies whose product launches machines from versioned images. Each workload gets a full VM with its own credentials, so one prompt-injected agent cannot reach your keys or its neighbours.
What sets it apart: spend limits and access scopes are set per machine before launch, agents can self-serve within them, and idle machines are snapshotted and suspended automatically. Billing is by the minute, with no plan to learn.
- Tagline
- A safe place for every agent to run, and a cap on what it costs.
- Voice
- PlainCalmPrecise
- Why this name
- Agents perch on a machine, stay put, and fly off when done. A perch line is also a row of them: a fleet. Short, easy to say, easy to type.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
One-line machine launch
Create, stop, snapshot and delete machines from a CLI with self-explaining commands, so a person or an agent can use it with no docs.
No member yet
Agent self-serve with budgets
Agents request their own machines through an agent protocol server, inside a budget, machine count and size limit you set per project.
No member yet
One VM per workload
Every agent gets a full VM with its own scoped credentials. Nothing shared, nothing inherited from your personal machine.
No member yet
Auto-snapshot and suspend
Idle machines are snapshotted and suspended on your rule, then resume where they stopped. You pay only for the minutes used.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#1F6F5CSpruce · primary#F6F1E7Oat · surface#1E2A28Charcoal Pine · text#E8833AEmber · accent#4A6D8CSlate Sky · data
Typography
Aa
Perchline
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
A safe place for every agent to run, and a cap on what it costs.
PricingAI proposal
The material gives only a public claim that machine0 starts at $0.013/hr and scales to H100/H200 GPUs. Machine usage here is billed by the minute at cost plus margin; the monthly tiers are a starting point to test.
Pay as you go
Free
- Per-minute machine billing
- CLI and agent protocol access
- Spend caps and auto-suspend
- Up to 5 running machines
Team
$29 a month
- Everything in Pay as you go
- Unlimited machines
- Versioned images and rollback
- Shared projects and approvals
- Alerts to phone and email
Scale
$199 a month
- Everything in Team
- Per-agent audit log
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
- Partners and integrations. The agent protocol server and CLI are listed in agent frameworks, coding-agent setup guides and example repos, so agents that need compute pick Perchline as their default way to get a machine.
- Do they become customers?
- People who find it through a framework are the developers who set it up, so they can become paying customers. To test: whether framework users who try a free machine go on to run a paid project.
- First customers
- Small ML and agent-tool teams of 3 to 15 engineers who already run several coding or training agents and have had a laptop sleep or a credential scare.
- Where to reach them
- Hacker News and dev forums: Post a write-up of an agent fleet with per-machine budgets, with a free starter credit and a repeatable demo.
- Agent framework communities: Offer a ready MCP config and example repo in Discords and GitHub discussions of popular agent tools.
- Direct outreach to ML teams: Message teams posting about multi-day RL or training runs and offer a free trial of a GPU machine.
- First milestone
- Ten teams each run a paid agent fleet for four weeks, with at least three spending over $200 a month.
Why people already pay for this
- VerifiedHacker News post received 83 points and 44 comments
- Public claimAgent workloads are moving from ephemeral to sustained compute, with coding agents working 6-8 hours and training/RL runs taking days.
- AI inferenceMultiple agents on a shared machine saturate resources, and there is acute demand for GPU access that personal infrastructure cannot meet.
- Public claimRunning agents on personal machines creates security exposure to credential exfiltration via prompt injection.
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 · Machine supplyKVM VM and GPU capacity via provider APIsFull VMs with real GPU drivers need a host provider; the product orchestrates them rather than owning hardware at first.
- This app · CLICross-platform CLI published on npmThe main way people and agents launch and manage machines.
RisksAI inference
- CommercialReselling provider capacity leaves thin margins against large clouds and the product named in the material; price and trust must be won on simplicity.
- TechnicalReliable snapshot, suspend and resume of VMs, especially with GPUs attached, is hard and failures lose a customer's multi-day work.
- PlatformDepends on upstream providers for GPU supply and API stability; shortages or terms changes would hit launches directly.
- CommercialDemand evidence is a single launch post with public claims from the competitor itself; real willingness to pay is unproven.
- RegulatoryRunning arbitrary customer workloads brings abuse, fraud and acceptable-use duties, plus data location questions for business buyers.
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.