The platform
Agents need a platform. You should own it.
A model gateway, retrieval over your systems of record, evaluation, audit and identity, stood up inside your environment in the first three weeks. It is yours whether or not you ever build another agent with us.
What gets stood up.
Five layers, all inside your environment. Each one is the answer to a question a controller or a CIO asks before an agent is allowed near a ledger.
Layer 01
Model gateway
One route to frontier and open-weight models, with the choice of model as a configuration rather than a rewrite. Spend, latency and failure are visible per workflow, so a model that gets slower or more expensive is something you notice rather than something you discover in an invoice.
Swap a model without touching an agent.
Layer 02
Retrieval over your systems of record
Grounding against SAP, NetSuite or Dynamics, plus the documents and mailboxes where the real exceptions live. The point is not a chatbot over a wiki. It is an agent that can read the same ledger a controller reads, with the same permissions.
Your data stays in your environment.
Layer 03
Evaluation harnesses
Every agent ships with a suite that runs against real historic transactions, so a change is measured before it reaches production rather than after. This is the difference between an agent you can improve and one you are afraid to touch.
You can change an agent without guessing.
Layer 04
Audit logging
Every action an agent takes, what it read to decide, what it drafted, who released it and when. Written to be read by an auditor rather than a developer, because the controller who signs the close is the person who has to defend it.
The trail exists before anyone asks for it.
Layer 05
Identity and access
Agents authenticate as themselves against your directory, with scopes a human can read and revoke. An agent should never hold a shared service account, and nobody should have to reason about what it can reach by reading code.
Revoke an agent the way you revoke a person.
Rented infrastructure is a fourth vendor.
The market has split into companies that sell you the foundation and companies that sell you agents running on a foundation they keep. Both leave you renting the part that should be an asset.
It runs in your environment
Your cloud, your accounts, your network boundary. Not a tenant on ours, and nothing calls home for permission to work.
It survives us
The gateway, the retrieval layer, the harnesses and the logs are all still there after the engagement ends. So is the ability to point them at an agent we had nothing to do with.
It is one thing, not two contracts
Buying the foundation from one vendor and the agents from another leaves you owning the integration between them. That integration is where these programs stall.
Stood up in three weeks, during Anchor.
The platform is not a separate project with its own business case. It is stage two of the same engagement, and the first agent goes live on top of it inside 90 days.
Assess maps the workflows and produces the ranked list. Anchor stands up everything on this page. The first agent goes live behind an approval gate. Then more agents, then the handover.
Bring us one workflow.
Thirty minutes, no deck. Pick the process that costs you most in people, errors or delay, and we'll pressure-test whether an agent is genuinely the right answer for it.
- An honest read on whether that workflow is a good agent candidate
- What the first deployment would involve, and roughly what it costs
- An honest answer if the sequencing is wrong and you should fix something else first