SJV Labs • In development

Meet the idea behind SJV Relay.

Relay is the working name for an internal platform SJV is developing to coordinate AI-assisted work across business systems. It is intended to control what agents can access, where people must approve an action, how execution occurs, and what evidence is retained afterward.

Working product name. Not yet offered as a standalone commercial SaaS product.
Product thesis

The model is not the operating system.

Models will continue to change. The durable value sits around them: governed access to systems, persistent job state, business rules, approval paths, reliable execution, evaluation, and an audit history that shows what happened and why.

Control

Policy & permissions

Role-based tools, tenant boundaries, secret isolation, risk-based approvals, and deterministic execution paths.

Orchestration

Agents & workflows

One orchestrator coordinating specialist agents for research, security, engineering, documentation, sales, and operations.

Evidence

Audit & evaluation

Jobs, artifacts, tool calls, approvals, outcomes, cost, reliability, and repeatable evaluations captured in one place.

How Relay is intended to work

One request. Controlled execution across the stack.

Relay is being designed as the layer that turns a business request into coordinated work, policy checks, approvals, and traceable actions.

01 Request Business event, task, alert, or goal
02 Orchestrator Plans work and delegates specialist tasks
03 Specialist agents Research, security, engineering, QA, documentation
04 Policy + approval Permissions, risk gates, human review
05 Business systems Microsoft, SaaS, APIs, data, workflows
Build in public, carefully

SJV will operate on the system before asking customers to trust it.

The internal lab is the proving ground: research, competitive intelligence, prospecting, proposals, software development, code review, QA, documentation, knowledge management, and controlled Microsoft operations.

Relay architecture Lab roadmap
Orchestrator Receives work, delegates, reconciles outputs
Core
Specialist agents Research • security • engineering • QA • documentation
Building
Approval & policy engine Human control over consequential actions
Building