Teams come to Liskov for three reasons: where the code runs, who it depends on, and what it costs. Acurast supplies hardware-backed execution on independent phones, each on its own home broadband. Liskov turns that into Applications you can release, operate, and prove.
Every cloud sells the same thing: a slice of a building you do not own, behind an address block everyone recognizes, on one provider’s terms. Liskov Applications run somewhere else.
A probe in a datacenter sees a datacenter’s internet: clean routes, a recognized address block, and none of the CDN, DNS, ISP, and geo-targeting behavior your customers get. Acurast processors are phones on home broadband. From there you measure the real thing.
Build now01
Edge monitoring and observability
Uptime, latency, DNS, TLS, and page-load checks on a schedule, from residential circuits on many ISPs in many countries. Catch the regional outage, the misrouted CDN edge, or the broken anycast node before the support tickets arrive.
For exampleYour status page is green from us-east-1, but customers on one national ISP have been timing out for twenty minutes. A probe on that ISP sees it. A probe in a datacenter never will.
Synthetic monitoring · Last-mile latency · DNS and CDN checks
Interval execution
Residential vantage
Parallel jobs
Country placement
Build now02
Real-device browser testing
Headless Chromium on real ARM phone hardware, behind a real residential IP. Render, screenshot, and click through your site the way a handset does, not the way a cloud emulator pretends to.
For exampleVisual regression on every release, rendered on the devices and networks your mobile users actually hold, with screenshots and timings posted back to your pipeline.
Puppeteer · Screenshot pipelines · Core Web Vitals in the field
Phone hardware
Residential vantage
Once or interval
Encrypted code
Build now03
Web intelligence and market data
Price monitoring, stock checks, search and listing collection, and ad verification from the vantage a real consumer has. A residential circuit gets the page people see, not the page served to a datacenter.
For exampleEvery hour, confirm that the price, stock, and promotion a partner shows in one market match what they agreed, with the evidence attached.
Scraping · Price monitoring · Ad and SEO verification
Interval execution
Residential vantage
Managed secrets
Country placement
Roadmap04
Regional experience verification
Choose the countries a job may run in, then confirm what each market actually receives: localization, geo-pricing, content availability, consent banners, and regulatory notices, measured from inside the market.
For exampleShow a regulator or a licensing partner that a stream plays in the territories you hold rights for, and nowhere else, from real devices in each one.
Geo-compliance · Localization QA · Rights enforcement
Country placement
Spread by country
Interval execution
Proof chain
02 — Decentralization
Nothing to seize. Nothing to switch off.
A cloud region goes dark. A conflict cuts a country off. A sanctions ruling closes an account overnight. In each case one building or one decision takes a whole service with it. A Liskov Application has no single building and no single decision-maker.
One conventional deployment
A list of single points of failure
One region. A power, network, or control-plane fault takes every instance in it.
One provider. Their outage, their price change, their policy change. Your downtime.
One account. A billing dispute, a compliance flag, or a sanctions ruling, and the console goes read-only.
One jurisdiction. A court order or an export rule where the datacenter stands reaches everything inside it.
One address block. Known, listed, and blocked as a class.
One Liskov Application
Independent devices, independent owners
Many processors. Each job runs on its own device. Losing one costs a replacement, never the Application.
Many operators. Processors belong to independent people and companies on an open market, not to one vendor.
Many networks. Residential ISPs rather than one backbone and one recognizable address block.
Many countries. The fleet spans countries, not one provider’s region list.
One durable record. Liskov keeps the release, policy, deployments, and evidence together, so the Application outlives any job.
Open-market assignment spreads jobs across independent processors today. Guaranteed spread across operators and countries is on the roadmap.
Build now05
Services that outlive a provider
Workers, schedulers, watchers, and notifiers that have to keep running when a region or a provider does not. When a job is lost, Liskov replaces it on another processor and keeps the Application’s posture, history, and spend in one place.
For exampleA payment reconciliation worker that cannot stop because a hyperscaler region in the Middle East is unreachable, and must not depend on one cloud account staying in good standing.
Keepers, oracles, indexers, relayers, bots, and validator monitoring on hardware with no single address, no single owner, and no single off switch. Execution is attested, and secrets are released only to the intended job.
For exampleA liquidation keeper or an oracle feed whose keys never leave hardware-backed execution, and which does not stop because one VPS provider pulled a plug.
Keepers · Oracles · Indexers · Telegram bots
Encrypted code
Managed secrets
Continuous execution
Public ingress
Build now07
Private computation on data that cannot go to a cloud
Code and inputs are encrypted for the assigned job and decrypted only inside the processor’s hardware-backed boundary. The owner of the phone cannot read them.
For exampleSigning, scoring, or matching over material whose residency rules rule out a shared cloud tenant, with the proof chain showing which release ran where.
Keep a workload inside the countries it must stay in, or out of the countries it must never touch, and spread it so no one country holds every job. The rule is part of the release contract, so the evidence shows it was honored.
For exampleA workload that may not run under one country’s legal reach, placed across processors in three others, with the placement on record for the auditor.
Data residency · Sanctions compliance · Sovereign deployments
Country allow and exclude
Spread by country
Proof chain
03 — Cost
Pay for the work. Not for the waiting.
The hardware already exists, already has power, and is already online. Acurast lets it earn by doing your work. Liskov quotes a job before it spends, reserves the credits, and settles the real charge, so the bill reads in finished work, not idle hours.
Build now09
Scale-out batch and media
Image and video conversion, document processing, WebAssembly pipelines, and any job that splits into independent pieces. Fan out across as many processors as the queue needs and pay for the jobs that ran.
For exampleTen thousand HEIC uploads converted overnight across a hundred jobs, with no transcoding fleet to size, patch, or leave running on Monday.
Video transcode · HEIC to PNG · WebAssembly · Benchmarks
Once execution
Parallel jobs
Native-image runtime
Quoted before spend
Build now10
Workers and bots instead of a VPS fleet
Cron jobs, fetch-transform-deliver pipelines, notification bots, and small daemons that each used to justify a virtual server of its own. Run them as Applications, draw on one organization-wide pool of job slots, and retire the fleet.
For exampleTwelve small services on twelve small servers become twelve Applications on one pool, each with its release, logs, and evidence in one console.
Assistants and local model services that hold API keys, credentials, and memory. Run them on inexpensive ARM hardware, release secrets only to the job, and reach them over managed Runtime SSH when they need a hand.
For exampleAn operations agent that watches your systems and acts on them around the clock, with no GPU bill and no credentials sitting on a shared cloud host.
OpenClaw · Hermes · Local LLM services
Continuous execution
Runtime SSH
Managed secrets
Durable state
Roadmap12
An inference swarm, priced by result
Batch inference, synthetic data, evaluation, and embeddings, with a complete small open-weight model on every worker. Throughput grows with the fleet, not with a purchase order, and a lost worker costs one retry.
For exampleClassify a backlog of records with a small open-weight model spread across the swarm, and read the bill in records classified.