Skip to content
CodeNSM LIVE company_mode=true

// FOR AI-NATIVE ENGINEERING TEAMS

CodeNSM is tracking CodeNSM · live

Stop shipping AI-generated code nobody understands.

See which parts of your codebase are load-bearing, fragile, or quietly rotting — and who owns them — without reading every diff. CodeNSM turns every function into a code employee with a role and a health state, so you catch AI-generated technical debt before it pages you.

One pip install, names-only by default — your source never leaves your machine. Automated Tech Lead judgment for the LLM-coding era.

// takes 10 mins · no source upload · see your risk profile immediately · no card · cancel anytime

CISO-Ready · 0% source exposure (names-only mode)

  • 100/100North-Star Metric
  • 786Code employees tracked
  • 12Stressed paths surfaced
  • 0Raw source files transmitted

// live proof

This is CodeNSM reading its own code, right now.

Every number on this page is CodeNSM running its own SDK on its own production codebase — so before you trust it on your code, you can watch it work on ours. Public proof, private architecture.

PUBLIC-SAFE FEED

Workspace

Confirmed live

Last scan 2s ago

Staff roster

786 employees

5 archetypes on the floor

Health

69% Fit

6% stressed · 18% snoozing · 5% external

Architecture drift

Down 2.3%

7-day trend · improving

Privacy boundary

Source transmitted 0

Derived signals only

Recent safe event

Trust-path stress detected

Details redacted

Live telemetry from CodeNSM's own workspace — aggregated, rounded, delayed & redacted. transmit_raw_source=false

// the problem

Your AI writes code faster than anyone can review it.

Velocity is up. Architectural clarity is down. Modules land in minutes, but nobody can tell load-bearing from disposable — or which AI-written function is a fragile trust-path liability. The gap hits two people differently.

If you're the founder

You're flying blind.

You're shipping features and hitting your roadmap — but you can't see the structural integrity underneath. You have no way to judge whether the codebase is getting stronger or more brittle, so you wait for an incident to tell you that debt quietly became a liability.

If you're the engineer

You're maintaining code you didn't write.

You're responsible for a growing mountain of AI-generated functions you never authored — impossible to hold in your head, painful to debug, and slow to audit. When something breaks, you don't know which path carries the risk until you're already in it.

// the tools you have show files, folders and diffs — not which function is fragile, load-bearing, or rotting.

// cost of inaction

Every sprint you don't map, drift compounds into rewrites.

  1. 01A reachable refactor this month becomes a full rewrite next quarter.
  2. 02A fragile trust-path worker ships to prod unreviewed — because nothing flagged it.
  3. 03The codebase gets harder to reason about exactly as you scale spend on more AI code.

// the mechanism

See the signal, keep the source.

Most tools force a bad trade: name-only scanners see no structure, and deep analyzers demand your whole repo. CodeNSM does neither. It derives structure locally with the standard-library AST and transmits only names + derived signals — a digital X-ray for your AI architecture that reads behavioral health, never raw source. We map every function so that you can see the brittle path so that it never pages you on a weekend.

structural preview approximates the local SDK's AST pass — not the exact production classifier

What CodeNSM derives and transmits

alias auth.fn_•••7a
archetype Security & Compliance
health state Stressed

Elevated branching + retry pressure on a trust path. Review first.

branching complexity88%
calls 8 branches 7 external no retry detected functions 1

What never leaves your machine

This ran entirely in your browser. In names-only mode only the derived signals above — never your source — would be transmitted.

Traditional tools show

  • Files
  • Folders
  • Commits
  • Lines of code

CodeNSM shows

  • Roles
  • Workload
  • Value paths
  • Health states
  • Review priority

// source-safe by design

Built for teams that cannot casually upload their source.

Two SDK depths keep your source on your machine. One optional connector — GitHub — is the single place you can grant code access, and only if you want commit-level ROI. You choose the line.

Names-Only Mode · recommended default

Zero code leaves your machine.

Allowed off-machine

  • Function names
  • Call / error / latency counters

Blocked

  • Source code
  • Arguments & return values
  • File contents

local → derived signals → dashboard

Local Full-Code Mode

Deeper analysis. Still local.

Control-flow shape is extracted on your machine with the standard-library ast. Only the shape — decisions, loops, call edges — ships. Never the source.

Stays local

  • All source text
  • The AST parse itself

mode="full" · local AST only

Connect GitHub · optional, you grant it

The only door to your code.

Everything above is derived on your machine — no repo access. Link GitHub and you give CodeNSM read access to your repository, so it can join commits to functions and price engineering time (commit-level ROI). This is the one place code leaves your world, and only because you opened the door.

You’re granting

  • Repo read access (commits, diffs, patches)

Unlocks

  • Commit-wise check-in view
  • Money-chart ROI & grade

optional · revoke in GitHub anytime

Through the SDK, your source never leaves your infrastructure — names-only or local AST. The only way CodeNSM sees code is if you connect GitHub for commit-level ROI. Your call, reversible anytime.

CISO-Ready · 0% source exposure (names-only mode) — the artifact your security team needs for a yes.

RoundedDelayedRedacted transmit_raw_source=false

// how it works

From one install to your first fixed risk — in one sitting.

  1. INSTALL · 1 MINRun the SDK in your local environment.One install, one init call. Names-only by default — nothing to clear with security.
  2. SCAN · 3 MINCodeNSM finds your load-bearing functions.The floor lights up: which code carries the traffic, the money, and the reach when it breaks.
  3. TRIAGE · 6 MINSee the 3 paths most at risk of failing.Fragile, overloaded and smell-ridden functions ranked by what a failure would cost.
  4. FIX · DONEShip the AI-ready patch.Copy the ticket-formatted fix prompt into Claude, Cursor or Jira — evidence attached.
$ pip install https://thinknorth.consulting/static/codensm/dist/codensm-1.0.0-py3-none-any.whl
# then, at startup:
>>> codensm.init(ingest_key="…", endpoint="…/codensm/ingest")
>>> codensm.autowatch("myapp")

Python shown. Node, PHP and Java ship the same five-outcome protocol with their own idiomatic SDK — codensm scan in CI, auto-instrumentation, or a decorator. Every way to wire each language in →

// takes 10 mins · no source upload · see your risk profile immediately · cancel anytime

// more proof · production

codensm/self · confirmed live

100/100NSM
786Employees
12Stressed
RedactedArchitecture
0Source transmitted
2sLast scan

The demo is CodeNSM watching its own workspace.

The same product surface used everywhere on this page is generated from CodeNSM running its own SDK on its own production code. If the product works, you are looking at it working — on itself.

Open Public-Safe Live Workspace →

// view-only · public-safe · architecture-redacted · no sign-up

// more proof · the aha

Your functions become a staff roster you can actually judge.

Select an employee — the inspector explains why it matters and what to do. This is the same product surface, running public-safe. We rank by risk so that your attention lands on the function that carries it.

EmployeeArchetypeStateWorkloadLatencyValue PathNext Action
billing.fn_•••2cSpecialist EngineerTop PerformerHighFastRevenueProtect
auth.fn_•••7aSecurity & ComplianceStressedMediumElevatedTrustReview
notify.fn_•••4fExternal LiaisonNetwork LatencyHighSlowCommsCheck retries
parse.fn_•••91Records ClerkSnoozingLowFastOpsLeave
release.fn_•••a8Floor SupervisorFitMediumNormalReleaseMonitor

// more proof · the platform

One product surface. Judgment for founders, engineers, and AI-native teams.

office.map() Virtual Office

The floor plan of your company — glowing employees, connector lines between value paths, aliases redacted. Watch load move in real time.

nsm.trace() North-Star Metric

100/100

Revenue · Trust · Ops · Release

staff.roster() Employee Roster

EmployeeStatePriority
billing.fn_•••2cTopProtect
auth.fn_•••7aStressedReview
parse.fn_•••91SnoozingLeave

flow.render() Flow View

Node-link map of a request. Stressed node flags amber.

commit.align() GitHub Alignment

Link PR clusters to the value paths they actually moved.

fix.prompt() AI-Ready Fix Prompts

A repair prompt for your worst function — paste straight into your agent.

standup.check() Cadence

The standup, made falsifiable. Every prediction in the room checked against the commit log and the telemetry — not against who sounded most certain. Blockers, unplanned work, and each developer's calibration by area, with the denominators.

sdk.any() Four Languages

One five-outcome protocol, four first-party SDKs. A test that throws isn't an outage; a declared 404 isn't a failure — only a real escape counts.

Founder View

A codebase health dashboard you can read

NSM 100Revenue Fit Trust StressedBoilerplate 18% Source 0
  • Judge engineering without reading code — one number, one roster.
  • See where structural risk is accumulating, not just what shipped.
  • Your one stressed worker, surfaced in 10 minutes.
Open founder view →

Tech Lead View

The daily walk-the-floor, automated

ALERT · trust path

auth.fn_•••7a crossed retry-pressure threshold. Debt trajectory ↑.

  • Modules ranked by debt trajectory — fragile-and-load-bearing first.
  • A pre-scrum brief generated from your commits and telemetry — before the meeting.
  • Every standup prediction checked against what actually shipped.
Open tech-lead view →

AI-Native Safety View

A safety net for code you didn't hand-write

generated module · review

2 brittle · 4 solid paths

  • Know which AI-generated modules are brittle before they bite.
  • An AI-ready fix prompt for your worst function on day one.
  • Flow diagrams for the code your prompts produced.
Read the SDK docs →

// access paths

Choose the safest way to see CodeNSM working.

Start Free

For teams ready to instrument their own app.

  • Names-only mode by default
  • Live staff roster + NSM
  • AI-ready fix prompts
Get Started Free →

no card · 10-min setup

Watch Live Demo

For anyone who wants to see it before installing.

  • CodeNSM watching itself
  • Public-safe, redacted
  • No sign-up
Watch Live Demo →

no sign-up required

Join Product Updates

For teams tracking the roadmap.

  • Telemetry digest
  • New capabilities
  • No spam
Join Updates →

unsubscribe anytime

// no pricing commitments shown

// nothing to lose by looking

Everything about this is reversible except not looking.

  • 🔒 Source-safeNames-only by default. Your raw source never leaves your infrastructure — transmit_raw_source=false.
  • 💳 No cardStart free. No credit card, no sales call, no commitment to see it working.
  • 🧩 Stdlib-only pathThe SDK runs on the Python standard library. Nothing exotic to vet before you install.
  • ⏱ 10-minute setupPip install to confirmed-live in about ten minutes. If it's not useful, uninstall — cancel anytime.

// the questions engineering teams ask first

The vocabulary, and the questions.

Code employee
A single function, treated as a worker with a role, workload and health state.
Staff roster
The full directory of your functions with archetype, state and next action.
Names-only mode
Default depth: function names and counters leave your machine — zero source.
Architecture drift
Movement of the North-Star Metric over time as AI-generated debt accrues or clears.
Stressed path
A load-bearing function under elevated branching, error or retry pressure.
Load-bearing function
Code that carries real traffic on a value path — expensive to break.
AI-ready fix prompt
A deterministic repair prompt for one diagnosed function, ready for your agent.
North-Star Metric
One importance-weighted health score per project — the number that summarizes the floor.
The five outcomes
Every call is ok, signal, handled, propagated or fault — and only a fault counts against you. A test that throws on purpose is not an outage; a declared 404 is not a failure.
Cadence
The standup made falsifiable: every prediction in the room resolved against the commit log and the telemetry, so calibration is measured from what came true — never from what was said.
Does it upload my source?

No. In names-only mode, only function names and counters leave your machine. In local full-code mode, control-flow shape is extracted locally via the standard-library AST and only that shape ships. Your raw source never leaves your infrastructure.

How does CodeNSM help with AI-generated technical debt?

It classifies every function — including the ones your AI wrote — into an archetype and a live health state, then flags the fragile, load-bearing paths first. You see which AI-generated functions are brittle before they bite, so you can manage AI technical debt instead of meeting it during an incident.

Is CodeNSM suitable for AI-first founders?

Yes. The Founder view is a codebase health dashboard for non-technical founders: one North-Star Metric and a staff roster you can read without opening a file — so you can judge engineering and see where structural risk is accumulating without reading code.

Will it slow my team down?

No. CodeNSM points your team at the few functions that carry the most risk instead of every diff, and the SDK only records lightweight names and counters — no heavy runtime overhead. It's built to save review time, not add process.

Does it work with AI-generated code?

That's exactly what it's for. CodeNSM doesn't care who wrote a function — it observes how each one behaves and how complex it is, which is precisely the visibility teams lose when AI ships code faster than anyone can reason about it.

What happens after I start?

You pip install the SDK, add one init call and choose a mode. Within about ten minutes your staff roster and North-Star Metric go live, your first stressed path is surfaced, and you get an AI-ready fix prompt for your worst function. No card, cancel anytime.

// see your own codebase

See which of your functions is fragile — before it pages you.

Start free in about ten minutes, or join the telemetry digest if you're not ready to install. Either way, your source stays exactly where it is.

takes 10 mins · no source upload · see your risk profile immediately · no card · cancel anytime

// no spam · or just get started now — it's free.

Start Free → Live Demo