Managed Agent Workstations

Business plan, product spec, market research and business model

Prepared by OctoMonic · November 2026 · Internal, not for external circulation

Executive summary

OctoMonic already runs one of these for itself. Operator is an always-on AI agent on a cloud VM, commanded over WhatsApp, that does real work for the business. The plan is to package that same thing and sell it as a done-for-you product: a dedicated AI operator per client, with the data pipeline built, hosted and maintained for them.

Two things make it valuable to a client. First, they get a capable AI agent doing ops or data work without hiring an engineer or learning any tools. They just message it. Second, where the work is legitimate web data collection (price monitoring, ad verification, QA), we build in a dedicated, vetted egress path (explained below) so their collection is reliable and does not get blocked.

The constraints confirmed by this research shape the whole plan:

• The subscription-vs-API saving is not ours to resell. Anthropic, OpenAI and Google all forbid pooling or reselling subscription accounts. Each client uses their own account. Our margin comes from hosting, building and running the pipeline, not from LLM pricing.
• The egress layer carries real regulatory weight. It only works as a vetted, low-volume, named-legitimate-use component with business-grade SIMs and client contracts, never as an open anonymity product.
• The managed layer is a genuine gap. No competitor bundles messaging control, a hosted always-on agent and a managed data pipeline. That is where we win.

Product and what we sell

We are not selling compute, and we are not selling an AI subscription. We are selling a built and operated outcome: "your competitor prices in your inbox every morning," or "your ads checked across every market daily," delivered by an agent the client talks to in plain language.

What an egress path is

Egress simply means outbound traffic: anything the agent's VM sends out to the internet. When the agent opens a website, that request leaves the VM through some network connection, and the website sees that connection as the visitor. The egress path is the route that outbound traffic takes, and it decides what the outside world sees: which network, which country, and what kind of connection (a data centre, a home line or a mobile network).

By default a cloud VM goes out through its data centre's connection. For many jobs that is fine. For web data collection it often is not, for two ordinary reasons:

• Accuracy. Many sites show different prices, ads and content depending on where the visitor is and what network they are on. A retailer may show mobile shoppers different prices than desktop visitors, or show Israeli visitors different ads than visitors from abroad. To report what real customers actually see, the request has to come from the same kind of network, in the same market.
• Reliability. Many sites treat traffic from data centres with suspicion and slow it down or refuse it, even for routine, legitimate monitoring. A pipeline that breaks every few days is worthless to the client.

So for data-collection clients we give the agent its own dedicated outbound route, set up for that client's job. Where the job needs it, that route runs over a mobile data connection on a SIM we manage, so the agent sees the site the way a mobile customer in that market would.

"Vetted" is what makes it a responsible product, not a loophole. A vetted egress path means:

• Vetted client: the client's identity is verified and they sign an acceptable-use agreement before the route exists.
• Vetted purpose: the route is provisioned for a named, legitimate job (for example, "monitor competitor prices on these sites"), and used only for that.
• Vetted connectivity: it runs on business-grade SIMs or wholesale agreements whose terms allow this use, not on consumer plans.
• Accountable: every request is logged per client, and we can shut the route off immediately if it is misused.

In short: the egress path is how the agent reaches the web. "Vetted" means we know who is using it, for what, on what connection, with a record of everything it does.

Product spec

Architecture

• Per-client VM. Isolated, hardened box, one tenant per VM, no shared state between clients. Sized per workload.
• Agent runtime. The AI agent runs headless on the VM, authenticated with the client's own credentials, inside the client's own workspace.
• Command bridge. WhatsApp, Slack or Telegram inbound to the agent, and replies, approvals and reports back through the same channel. Modelled directly on Operator's existing bridge.
• Egress path. For data-collection clients, a dedicated outbound route (mobile-grade where the target requires it), provisioned per client and used only for the client's named, legitimate collection job.
• Approval gates. High-impact or outbound actions pause for a human yes, the same gate pattern OctoMonic uses internally.
• Audit log. Every command, action and egress request is logged per client, for the client's compliance and ours.

Service tiers

What the ongoing service covers

• Hosting, uptime and monitoring of the VM and egress path.
• Keeping the pipeline working as target sites and tools change. This is the real ongoing labour.
• Agent tuning, new instructions and small scope additions.
• A regular report and a standing WhatsApp line for the client.

Larger scope changes are quoted as separate mini-builds, which protects the ongoing service margin.

Use cases

Market research

Market sizes, competitor prices and dated statistics are in the numbers section at the end. The full sourced report is archived in AIOS.

Market

• The proxy and web-data market is growing steadily, and vendor revenue is growing much faster than analyst reports suggest. Demand from AI companies for web data is the main driver.
• E-commerce price intelligence is the top use case. AI data collection is the fastest-growing one.

Pricing landscape

• Raw hosting and sandboxes are cheap and commoditised. None of the providers offer messaging control or a managed service.
• Shared proxy pools have collapsed in price over the past year. Not a business to enter.
• Dedicated ports are the only defensible proxy niche.

Trends

• AI-agent web traffic is rising sharply.
• Anti-bot defences are tightening (Cloudflare now blocks AI crawlers by default), which raises the value of a professionally run, reliable collection pipeline.
• The industry is moving up the stack from raw proxies to managed scraper services, the same direction this product sits in.

Competitive position

Business model

Revenue

• Build fee (one-off): scoping, vertical template, pipeline build, VM and egress provisioning, onboarding and vetting. Funds the work and filters for serious clients.
• Monthly retainer: the recurring base covering hosting, maintenance, tuning and support. This is the long-term value of the business.
• Expansion: mini-builds for new scopes, extra markets or sources, extra agents, tier upgrades.

Costs

• Infrastructure per client: VM hosting, egress, monitoring and tooling. Small relative to the retainer.
• LLM usage: none for us. It runs on the client's own account.
• Labour: the build itself, then ongoing maintenance. Maintenance is the real recurring cost, spread across many clients per operations engineer.

Why the economics work

• Each client is cash-positive from the start, because the build fee covers the build labour.
• The retainer is high-margin, because infrastructure is cheap and the LLM bill sits with the client.
• Vertical templates shrink build time with every repeat, which is what makes several clients a month deliverable.

Go-to-market

• Lead with the reference case. Package GPI/GoMobile as a named case study and the first vertical template (price intelligence).
• Sell verticals, not technology. "Competitor prices every morning" sells. "Hosted agent VM with egress" does not. One landing page and pitch per vertical.
• Start warm. A few warm leads in price intelligence and agencies, to validate pricing before any hiring.
• Template the build. Vertical templates are what make multiple clients a month deliverable.
• Land and expand. Start a client on one job or market, then grow through mini-builds and higher tiers.
• Channel. Agencies and consultancies that have the clients but not the pipeline capability, through white-label or referral.

Team and operations

Risk and compliance

Hard rules on the egress layer

• Vetted clients only. Identity verification and a signed acceptable-use agreement before provisioning. No exceptions, including at low volume.
• Named legitimate uses only. The contract lists what the pipeline is for. No open-ended anonymity service.
• Business-grade SIMs or wholesale agreements only. Not consumer plans. Their terms almost certainly forbid this use, and the cheap-consumer-plan margin is not a real business.
• Full audit logs per client, and the contractual right to cut a client off immediately.
• Low volume by design. This is a boutique posture. The control is the product.

Why the posture matters

• NetNut, a Nasdaq-listed Israeli proxy operator, had its network seized by the FBI and IRS in July 2026, paused services and made major staff cuts. "Israeli proxy operator" is a high-scrutiny label right now.
• Google took down the IPIDEA proxy network in January 2026. The UK made unjustified SIM-farm possession a criminal offence in October 2026. Europol and the US Secret Service ran large SIM-farm seizures in 2025.
• Industry leaders (Bright Data, Oxylabs) run human-reviewed KYC and use-case limits. Matching that standard is table stakes. Small scale does not lower the legal exposure. One bad client is enough.

AI provider compliance

• Each client authenticates with their own AI account. We never pool or resell subscriptions, which Anthropic, OpenAI and Google all explicitly ban.
• Grey areas to confirm in writing with Anthropic (through the Claude Partner Network) before scaling: whether a VM we administer holding a client's token counts as us storing it, whether several staff can command one agent on one seat, and whether an always-on agent fits "ordinary, individual usage."
• Our own Operator, on Tom's own plan for OctoMonic's own work, is compliant as is.

Gates before any client build

• A telecom lawyer's opinion on Israeli registration or licensing for the egress service.
• Written business-SIM or wholesale quotes from several carriers or MVNOs.
• Written confirmation from Anthropic on the grey areas above.

Milestones and next steps