CSTM-SW · AGTcstmAI^™ · Software

AI workflow agents with a person at every gate you choose.

cstmAI^™ Agents carry out multi-step work inside your systems: reading an incoming document, checking it against your records, preparing the next step, and acting only when a person approves or within limits you set.

Get a quote →See the whole stack

Fig. 01 · Pattern storage above the shop floor, Paterson, NJ1994

An agent is a model with tools and a procedure. The hard parts are at the edges: what it may touch, when it must stop and ask, and how you can see afterward exactly what it did. Agents routes every action through the policies you set in cstmAI Govern and records each step.

New agents start in suggest-only mode, where a person approves every action. Their authority widens only where the evaluation results justify it, one action type at a time.

At a glanceCSTM-SW · AGT
Actions	Read and write in connected systems through cstmAI Connect
Approvals	Per action type: approve, edit or reject
Limits	Scope, volume and value limits per agent
Trace	Every step, input and tool call recorded
Models	Open-weight models served by the cstmAI Platform

Where it sits

Its place in the stack.

Every module runs on the same platform and hardware, under the same governance.

L5 · Your apps

Custom agents & applicationsBuilt by our development team on the layers below

L4 · Modules

SearchAnswers grounded in your own documents, with citationsAgentsWorkflow agents that act inside your systems, with human approval stepsStudioFine-tuning and evaluation on your data

L3 · Control

GovernAudit logs, permissions, data redaction and usage policyConnectConnectors to ERP, CRM, document stores and data warehouses

L2 · Platform

PlatformThe on-premises AI stack: model serving, access control, monitoring

L1 · Hardware

DeskCSTM-D · REV ARackCSTM-R · REV AClusterCSTM-C · REV AEdgeCSTM-E · REV A

Controls

Useful because it is supervised.

01CSTM-SW

Approval gates

A person signs off on the actions you choose, inside the tools they already use.

02CSTM-SW

Scoped tools

Each agent gets only the systems and records its job needs.

03CSTM-SW

Full trace

Every decision is replayable: what it read, what it concluded, what it did.

04CSTM-SW

Earned autonomy

Authority widens per action type as the error rate proves it safe.

Machinist standing at a large industrial lathe amid rows of machine tools on a factory floor

Reel 02 · Lathe area of a machine shop, Paterson, NJ1994

How it's delivered

Four steps, each one signed off.

Map the procedure

Write down the steps, systems and exceptions with the people who do the work.

Suggest-only

The agent drafts every action; people approve, edit or reject.

Measure

Approval and correction rates become the scorecard.

Widen

Low-risk actions with clean records get automatic approval.

FAQ

Questions we hear first.

What kind of work suits an agent?

Repetitive, rule-shaped work with clear inputs: invoice matching, intake triage, order checks, ticket routing, report preparation. Work that rests on judgment keeps a person in the loop.

What if an agent gets something wrong?

In suggest-only mode a person catches it before anything happens. Every action is logged with its inputs, so mistakes are traceable, and each one becomes a new test case.

Is this the same as custom agent development?

Agents is the runtime. When a workflow needs custom logic, integrations or a purpose-built interface, our development team builds it on top.

Where to go next.

developmentCustom AI agents, built for youOur engineers design the agent around your workflow.softwareThe policies agents obeyPermissions, limits and audit logs.hardwareProduction hardware for agentsA Rack serves agents to the whole company.

Get a quote

Spec your system.

Tell us the models you want to run, how many people will use them and where the hardware should live. An engineer replies with a first configuration and the questions that decide the quote.

Get a quote →How we build

Form CSTM-Q · Quote only