G-Whizz Engine
Captures instructions, routes work through agents, applies policy and records exchanges, runs, events, traces, costs and approval requests.
AgentView
One governed AI product. Two defined functions.
G-Whizz runs the work. AgentView makes it visible: agents, exchanges, runs, approvals, cost burn, traces, tool use and project movement in one read-only management suite.
ModeRead-only
WritesApproval gated
SourceLabelled facts
Human truthICE!...OS
Operating proof path
01G-Whizz engine
02AgentView management suite
03Human / ICE!...OS
04Approval gates
05Postgres read model
AI systems do not fail only because they act badly. They fail because nobody can see what they are doing.
G-Whizz and AgentView are one product with two jobs. G-Whizz turns instruction into governed work. AgentView shows the operating picture before anything moves.
One product. Two parts. A hard boundary.
G-Whizz owns intake, routing, orchestration, policies and execution facts. AgentView owns visibility, inspection and approval surfaces. Human / ICE!...OS remains the human source of truth.
AgentView Management Suite
Reads the G-Whizz operating facts, visualises the system, labels evidence and prepares scoped approval handoffs.
The engine and management suite stay separate by design.
G-Whizz creates the facts. AgentView makes those facts inspectable.
Part 01
G-Whizz Engine
Turns instruction, signals and system events into governed exchanges and routed runs.
- Intake
- Routing
- Policy gates
- Execution facts
Part 02
AgentView Management Suite
Reads the operating facts and turns them into reviewable management surfaces.
Operating PulseExchanges, approvals, events and costs.
Agent MapDirectors, agents, advisers, tools and data systems.
Approval QueueScoped human review before writes or spend.
Trace SummariesPhoenix-linked diagnostics without payload duplication.
Tool / MCP UsageConnector use, latency, dry-run posture and failures.
Project MovementActivity, stale work, weak evidence and sync candidates.
This is a functioning prototype.
The current build is not a static concept. It already presents the control-plane shape: live-read posture, approval pressure, topology inspection, cost visibility and explicit human gates before anything writes, spends, sends or deploys.
Public-safe prototype
AgentView can be shown as a real interface while redacting live work detail from public screenshots.
Read before action
The surface is organised around inspection, approval and evidence rather than autonomous execution.
Management suite only
G-Whizz remains the engine. AgentView remains the management suite for review and control.
Next phases move it from prototype to operating product.
The next build lane is a cleaner contract between the G-Whizz engine and the AgentView management suite, backed by Postgres, explicit evidence and controlled approval paths.
Read model hardening.
Move exchanges, runs, approvals, events, costs, traces and tool facts into stable Postgres-backed contracts.
Topology and evidence views.
Make G-Whizz, directors, agents, advisers, MCP tools and data systems inspectable with health, staleness and selected-node context.
Approval and release controls.
Turn approval queues into scoped handoff packets for writes, spend, sends and deployments, with clear human review points.
Public and operating release.
Prepare a public project page, private operator view and controlled demo path that shows capability without exposing live work data.