THREE LENSES

Same stack. Three ways to read it.

Every product stands alone, but the stack organises around three concerns: building and operating agents, governing how they talk to the outside world, and grounding, verifying, and settling whatever they produce. Pick the lens that matches the job in front of you.

LENS · BUILD

Build & operate agents.

Everything you need to run an agent in production; the runtime, the host, the builder, and the environments. Start here if the job is standing up a useful agent for a business, a team, or a developer.

R1 runtime Heroa host CloudSwarm builder CodeR1 coding env Actium assistants Actium Studio CMS

How it composes: R1 runs inside Heroa. Heroa hosts scratch instances the agent deploys via heroa.deploy(). CloudSwarm picks skills, schedules runs, and wires integrations. CodeR1 gives coding agents a persistent environment. Actium packages recurring jobs as hirable assistants with a native Tasks inbox and CRM; Actium Studio is the edge-native CMS sibling that shares the specialist platform.

LENS · GOVERN

Govern traffic & usage.

Full control over every AI request your company makes; policy, budget, sovereignty, and PII handling evaluated inline before the model sees the payload. Start here if the job is making AI usage accountable and keeping regulated data where it belongs.

RelayOne policy + sovereignty RelayGate programmable middleware

How it composes: RelayOne enforces origin, destination, residency, and spend across four distinct surfaces (sovereignty, employee AI access, network reroute, identity + commerce); the same guarantees hold on-prem, sovereign, hybrid, or cloud, without a cloud-provider audit dependency. RelayGate is the binary that sits between your agent and the model provider; ContextWorkers run inline inside the request for credential injection, PII stripping, and policy, not in a queue hidden behind a gateway.

LENS · TRUST

Trust, research, transact.

Ground every answer in evidence, verify every output against a spec, and settle every commercial event on a signed chain. Start here if the job is making AI output safe to show a customer, or safe to hand to a regulator.

DeepTap knowledge engine Veritize verification + drift TrueCom commerce substrate

How it composes: DeepTap grounds agent answers with a compounding Fact Cache; three depth modes, a seven-layer retrieval pipeline, MCP and A2A protocols, ZDR by default. Veritize verifies that the output actually uses the evidence; structured evaluators first, LLM evaluators last, drift tracked across model and prompt versions. TrueCom signs the commercial event; six primitives, ten settlement rails, seven dispute states, one call shape. The rail is an argument, not a rebuild.

Each lens has its own deeper reading in the five canonical integration paths; those are the specific compositions the portfolio markets to specific buyer shapes.