NetworkAccess · Product Emergence Plan

Where is NetworkAccess in all of this?

Short answer: it's the first product off the line — and the line is now built. For the last stretch we've been forging the loom (the Vinxi kernel). None of it looked like NetworkAccess because a loom doesn't look like a garment. Here's how, week by week, the garment appears — and why every new screen, data model, and workflow after that takes days, not quarters.

What the team sees today

"Ports, verbs, adapters, a commit path… none of this looks like the app we sell."

Correct. That's the engine, not the car.

Why that's the good news

NetworkAccess = the kernel, dressed. The hard part — trustworthy state, ontology, actions, replay — is done once.

Every NetworkAccess feature from here is an assembly, not an invention.

This is the payoff of building a kernel first. Every request your PMs and sales will ever make — "add a screen," "track a new thing," "automate that step" — is one of these five moves. None of them require touching the engine. That's what "keep growing" means, concretely.

New screen
Surface
Declare a surface + a query. The OS shell renders it, routes to it, permissions it.
≈ an afternoon
New data model
Ontology doc
Install an ontology document — new entity types, links, value-types. No DB migration.
≈ hours · zero downtime
New workflow
Process
Compose Acts into a process. State transitions become governed, replayable steps.
≈ a day
New action
Act / Op
Add one Op. It's authorized, logged, undoable, and on the wire — automatically.
≈ hours
New intelligence
Agent / Job
An agent or background job whose tools are the same syscalls people use.
≈ days, reusable

Each week ships something a salesperson can put on screen and a customer can touch. Every headline is the outcome the business sees; the ◆ chips underneath are the kernel parts that made it cheap.

Week 0 · nowThe loom is built

The engine exists — and it already knows telecom

Kernel core, 14 ports, memory + Postgres backends, the ontology, governed Acts, a surface runtime, and a working fault → correlation → ticket engine are all in. Fiber/PON is already modelled.

kernel-core · commit + replayontology (fiber/PON)Acts · ActionStoresurfaces runtimeRCA engine (Milestone A)
The riskiest, least-visible work is behind us — not ahead.
Week 1First light

NetworkAccess opens — live network inventory

The app has a face. The OS shell becomes NetworkAccess; the first surfaces show real inventory — OLTs, ONTs, splitters, fiber routes — read straight from the ontology. A customer sees their network.

surface · launcherEntityQueryPortontology → screens
Proves the data model drives real screens with zero glue code.
Week 2Situational awareness

The NOC map — the network, live, with alarms

The hero screen. A geographic / topology map over the fiber graph, alarms lighting up in real time. This is the demo that sells NetworkAccess in the first 30 seconds.

EntityGraphPort · traverseAssertStore · seriesraise_alarmsurface Phase 3
NetworkAccess · NOC Console
OLT-12 SECTOR-7
Live alarms · 3
LOS · Sector-7 splitter 12:04:11
214 ONTs unreachable 12:04:12
High RX drop · PON-3 12:03:50
Week 3From noise to answer

RCA that resolves tickets — one cause, not 214 alarms

The "wow". The 214 alarms above collapse into one ticket, pinned to the true fault — the Sector-7 splitter — with a confidence score and a one-click Confirm. That click is a governed Act: logged, attributable, replayable.

correlation · pure recomputesuspected_root_cause linkticket ActsActionStorePort
NetworkAccess · Ticket #NA-4471
NA-4471 · Sector-7 outageTRIAGE · auto-opened
Suspected root cause · RCA engine
Splitter SPL-7A — signal loss at the PON subtree cut
confidence 0.86
Cuts mean-time-to-repair from a war-room hour to a single click.
Week 4Get ahead of it

Proactive & predictive — alerts before the outage

From reactive to predictive. Series data + derivations run as background jobs and forecast degradation — "this link will breach in ~6 days." Sales story flips from "we fix faster" to "we fix before the customer notices."

AssertStore · seriesderivationsMaterializationRunPortworker jobs
NetworkAccess · Predictive Feed
RISK PON-3 optical budget trending to breach · ~6 days
FORECAST Sector-4 uptake → capacity at 90% in 3 weeks
FORECAST ONT batch B-19 failure rate above cohort
Week 5The copilot

AI agent — ask the network, get an answer with evidence

An operator just asks. The agent's "tools" are the exact same syscalls the UI uses, so it can query, correlate, and draft a ticket — and every answer cites the evidence it read. No hallucinated network state.

@vinxi/agent-hostKernelClient toolsplaybooks (QuickJS)read + Act with approvals
NetworkAccess · Copilot
OperatorWhy is Sector 7 down, and what do we do?
CopilotSector-7 is dark because splitter SPL-7A lost signal at 12:04 alarm:SPL-7A:LOS. 214 ONTs sit downstream graph · pon_downstream. Root cause confidence 0.86. I've drafted ticket NA-4471 and dispatch for the Sector-7 field team — confirm to send.
Week 6Hands-free

Voice — the same copilot, spoken, for the field

For the technician up a pole. Voice front-end on the same agent. "Raise a ticket for OLT-12 and assign the north crew." Because the action is a governed Act, a spoken command is as auditable as a click.

agent-host + speechsurfaceActs (voice = attributable)
NetworkAccess · Field · Voice
🎙"Raise a ticket for OLT-12 and assign the north crew…"
CopilotDone — ticket NA-4488 opened, north crew notified. Want directions to the cabinet?
Week 7Run the work

Projects & field workflow — rollouts, provisioning, crews

Beyond faults. Provisioning drives, rollouts, and maintenance become projects — entities with stages, where every transition is an Act and every persona (planner, field engineer, System Owner) sees their slice.

ontology kind · projecttransitions = Actsrosters · entitlements
NetworkAccess · Rollout Board
Planned
Ward-9 FTTH1,200 homes
Sector-4 upliftcapacity
In field
Ward-3 splicecrew · north
OLT-12 swapNA-4488
Done
Ward-1 cutoververified
Week 8+Self-serve growth

The platform grows without us in the loop

The compounding week. A new customer needs a bespoke screen, a new asset type, a new approval flow? It's an install of a document + a surface declaration — shipped in a day, by a solutions engineer, without a kernel release. This is the moat: every subsequent product feature gets cheaper, not more expensive.

unified package manifestinstall-package Opregistry-driven surfaces
NetworkAccess stops being "a build" and becomes "a configuration."

Everything on the wish-list is an assembly of kernel parts that already exist or are one small addition away. Here's each one, what it's made of, and the week it becomes demo-able.

AI agents
Operators and field crews talk to the network in plain language; the agent acts with evidence and approvals.
agent-host · KernelClient tools · playbooks
Week 5
Predictive analytics
Series + derivations forecast failures and capacity — proactive, not reactive.
AssertStore · derivations · worker jobs
Week 4
Voice interaction
The copilot, spoken — hands-free for the field, fully auditable.
agent-host + speech · Acts
Week 6
RCA → ticket resolution
Alarm storms collapse to one cause with a confidence score and one-click confirm.
correlation · RCA verdicts · ticket Acts
Week 3 · engine live now
Project management
Rollouts, provisioning and crews as projects — governed stages, per-persona views.
ontology kind · transitions = Acts · rosters
Week 7
The one-line version for the room: "We spent the build so far making a machine that turns a data model into working screens, actions, and AI — safely and instantly. NetworkAccess is the first thing we feed it. In roughly eight weeks it goes from an inventory list to a predictive, voice-driven NOC with an AI copilot — and the ninth feature is cheaper than the eighth."