What we do, what we don't,
and where the line sits.
Most vendor comparisons end in a maybe. This page does not. If you are running an evaluation, take these rows as answers. Where a row says no, it means no.
In one line
You keep your systems of record. We run the work between them — with AI agents and your own people on the same queue, to the same standard, and the business outcome written into the contract.
Not a copilot. Not a chatbot. Not a platform you license and then have to staff.
Capabilities
Twenty rows. Every one of them a commitment we will repeat in a room with your procurement team.
| Capability | Answer | What it means here |
|---|---|---|
| Build and run AI agents against enterprise workflows | Yes | Agents work real queues, not demos. |
| Human approval before an agent acts | Yes | A supervision desk where people and agents share one queue, one standard, one performance bar. |
| Escalation to a named person | Yes | Escalation paths are part of the rulebook, not a fallback. |
| Company-specific business rules | Yes | Your rulebook. Your entities, thresholds, policies and approvals. |
| Rules editable without engineering | Yes | The operations lead who owns the number edits the rule directly. Versioned, approved, no code, no ticket. |
| Rules enforced, not requested | Yes | The state cannot change except by a permitted move. An unauthorised action is refused, logged, and the agent is handed the moves it may make. |
| Audit trail of every action | Yes | Every action, by whom, on what evidence. Append-only. It only ever grows. |
| Learning from live operation | Yes | Live context that expires. Corrections captured from a supervisor's decision. Reflections distilled overnight. |
| Learning you can inspect | Yes | Every entry activated by a named person and stored as readable text with its provenance. A sentence someone can check, not weights nobody can read. |
| Reads and writes your systems of record | Yes | CRM, ERP, ticketing, billing, warehouse. They stay authoritative. |
| Arrives with your industry already in it | Yes | Ontologies, lifecycle patterns and rulebook structure for your sector, before day one. |
| We operate the workflow, not just ship the software | Yes | Forward-deployed engineers get it live and stay. Banded capacity, never priced per head. |
| Contractual ownership of a business KPI | Yes | A measured baseline, an agreed target, and remediation at our cost if the number does not move. |
| Runs in our cloud | Yes | Multi-tenant SaaS. Start without capital expense. |
| Runs in your cloud | Yes | Single-tenant VPC on Google Cloud, Azure or AWS. |
| Runs inside your perimeter | Yes | On-premise, air-gapped. |
| You can end up owning it | Yes | A perpetual licence after implementation. Your team runs it, or we do. |
| Replaces your system of record | No | Deliberately. We sit on top of it. Your systems keep the records. |
| A copilot or an assistant | No | The work is the product, not the suggestion. |
| A governance tool you install beside the operation | No | Governance is how the operation runs here. It is not a product sold next to it. |
| A consultancy engagement with an end date | No | We don't leave a report behind. |
| A free pilot | No | Pilots are paid, time-boxed, and credited against year one. |
Where the line sits
Three limits, stated here rather than discovered in month four.
A person with direct credentials can still do what your system permits.
What is governed is every automated path. If someone changes a record straight in the ERP, that surfaces as our state disagreeing with the source. That is detection, not prevention — and any vendor telling you otherwise is describing a system you do not have.
How much of the rulebook arrives on day one depends on your sector.
In a templated sector most of it is there before we start. Outside one, less. We will tell you which you are in during the first conversation, not after the contract.
The outcome commitment is scoped to a metric agreed in advance.
A measured baseline and a named target — not a warranty on your whole P&L. The boundary is what makes it worth anything.
Deployment and data residency
Two questions, and you answer them independently: where it runs, and who runs it.
| Option | Detail |
|---|---|
| Multi-tenant SaaS, our cloud | Fastest start. No capital expense. |
| Single-tenant VPC, your cloud account | Google Cloud, Azure or AWS. |
| On-premise, air-gapped | Inside your perimeter, no external connectivity. |
| Option | Detail |
|---|---|
| We do | Forward-deployed engineers operate the workflow against the agreed metric. |
| Your team does | After implementation, under a perpetual licence. |
| Both | The common shape. We run it, your operations lead owns it. |
Everything built for you — workflows, integrations, business policies, escalation hierarchies, your data, your memory — is yours from day one, contractually, in your tenant, exportable in open formats. Source escrow with defined release triggers is standard on deployed contracts.
What it costs, in shape
Three components. Build — one-time, fixed price, and we carry the scope risk. Licence and operations — annual; the platform plus the capacity running the operation against the agreed metric. Usage — metered per business interaction, where an interaction is a complete business event, not a compute event.
Engagements start with a paid pilot. A fixed period. One measured outcome. Credited against year one.
What is already true
Eighteen months live in production on a consumer e-commerce operation.
Resolved without a person
Escalation rate, with no change of model
Cost to serve
Ops headcount on the flow
Eleven AI workflows running inside the business. Contract to first workflow live: four to six weeks in a templated sector. ISO 27001:2022. ISO 9001. Two provisional patents filed. Delaware C-Corporation.
If a row above is the one you care about, say which.
Thirty minutes about fit — not a demo, not an audit, not a diagnostic. Bring the row you are stuck on.
CygnusAlpha. We turn AI ambition into AI operations.