DKOD is a governed build pipeline: it takes the risky, ungoverned AI-built internal apps that employees have vibe-coded across an organization and rebuilds them safely, under governance, from approved templates. Discovery with dkod-signals shows you what you have; the pipeline gives those apps a safe path forward.

What is dkod-signals?

A single static Rust binary that IT admins push via MDM (Jamf, Intune, Kandji) to every macOS, Windows, and Linux device. It runs once at low priority, inventories the AI-built apps on the device, scores each one for risk, and writes one privacy-safe, metrics-only JSON report. It is available today.

What does the signals report contain?

Metrics only: how many AI-built apps were found, their risk signals (no git, no auth, secrets in .env, production data access), and how many qualify for a governed rebuild. No source code, no file contents, no transcripts. The report is written locally and mirrored to a required S3-compatible bucket in your own cloud account. Uploading to the dashboard is an additional, opt-in step.

Which AI coding tools can it detect?

dkod-signals recognizes the traces left behind by Claude Code, Codex, Cursor, and other coding agents, and uses them to find the apps those sessions produced.

What can the pipeline rebuild?

The first rebuild templates are scheduled-digest (a periodic script whose output goes to Slack, email, sheets, or a webhook), dashboard (read-only data dashboards), and internal-web-app (internal-only web apps and APIs, no payments). Apps outside this envelope are flagged with explicit blockers — for example, "writes to production DB" — so nothing is silently ignored.

Do we have to ban AI-built tools?

No — that's the point. Employees built these tools because they needed them. DKOD gives you the picture first, then a safe path: rebuild qualifying apps through a governed pipeline instead of banning them or ignoring them.

Autonomous governed harness · Now available

Unblock AI for your whole org.
Governed. Autonomous. Yours.

DKOD is an autonomous harness that lets agents build and run your internal apps by following your organization’s policies. You see every agent, skill, MCP server, hook, rule, and budget. You set them your way. Nothing leaves your infrastructure without consent.

Join early access Read the docs

Why teams fall behind

Nobody can see what AI already built. So nobody can say yes.

Your people already build with AI: a revenue digest here, a customer dashboard there. That energy is an asset. DKOD gives you the picture, sets the guardrails, and lets it run.

TodayWith DKOD

No visibility.Every AI-built app, listed.
No trust.Policies you set. Gates that hold.
No way to say yes.A governed yes, with a reviewer where it matters.
Meanwhile, others move.Your teams move, with guardrails.

How it works

From shadow AI
to a governed harness.

Discovery is the wedge. The harness is the product. Every agent runs inside your infrastructure, under your policies.

01
See
dkod-signals · available today
One static Rust binary, pushed by MDM to every macOS, Windows, and Linux device. It runs once at low priority, inventories the AI-built apps from the traces coding agents leave behind, and writes one metrics-only JSON report. Now you know what you have.
02
Set policy
your rules become the gates
Approved templates, data boundaries, who approves what. Device reports roll up into one view of what qualifies, and your policies become deterministic gates that every agent action passes through.
03
Let it run
the governed harness · now available
Agents rebuild qualifying apps from approved templates and run them in your own cloud. Humans approve what matters. Every step is logged. Apps outside the envelope are routed to a person with a clear reason.

The harness

Agents do the work.
Your policies set the rules.

Instead of saying no to AI, you say yes with guardrails. Every agent action passes your gates. Humans step in only where you decided they should.

example run · illustrative

$dkod harness run --policy org/default

▸policy loaded: 14 rules · 3 templates · 6 skills · 4 mcp servers · 2 hooks

▸agent/plan revenue-digest → template: scheduled-digest · model: org-approved

✓gate skills allowed: digest, sheets-export

✓gate mcp allowed: slack, google-sheets · denied: prod-db

✓gate hooks ran: secret-scan, lint · no secrets in source

✓gate budget: 2.1k of 50k tokens · within team cap

▸agent/build revenue-digest → building in your cloud

⏳approval waiting for platform-team

✓approved by platform-team · audit trail written

✓running revenue-digest · scheduled daily 07:00

◼held customer-dash · writes to production DB → routed to platform-team

Policy as gates

01Agents and models
02Skills and MCP servers
03Hooks and rules
04Budgets and data access
05Approvals and audit trail

Deterministic. Logged. Yours to change.

Your control plane

See everything.
Control it your way.

Before DKOD, none of this was visible. It lived on laptops, in personal configs, in whatever each person set up. Now every one of these is something you can see, and set, for the whole org.

01
Agents
Which coding agents run, on which devices, and what they may touch.
allow: claude-code, codex, cursor
02
Skills
Which skills an agent can load. Approve them, version them, retire them.
skills: approved@v2 · retire: legacy-*
03
MCP servers
Which tools agents can reach. An allowlist per team, not a free-for-all.
allow: slack, sheets · deny: prod-db
04
Hooks
Actions that always run before and after an agent acts. Scan, lint, sign.
pre: secret-scan · post: lint, sign
05
Rules
The instructions every agent follows. Set once for the org, not per laptop.
rules: org/default → every repo
06
Permissions
Which commands and paths an agent may use. Everything else is denied.
deny: sudo, package installs · allow: bun test
07
Models
Which models, from which providers, for which kind of work.
models: org-approved · region: eu
08
Budgets
Token and spend caps per team, per project, per agent. Alerts before the wall.
cap: 50k tokens/day · alert: 80%
09
Secrets
No keys in .env files. Injected at runtime, scoped to the job, rotated.
inject: runtime · scope: job · rotate: 30d
10
Data access
Which datasets an agent may read. Production stays closed unless you open it.
read: analytics · prod: closed
11
Templates
The approved shapes an app can take. Everything else gets a reviewer.
scheduled-digest · dashboard · internal-web-app
12
Approvals and audit
Who says yes, for what, and a log of every action and every gate.
approve: platform-team · log: all

The envelope

Approved templates.
Clear reasons.

The first slice is the Digest Vertical. Three templates cover most of what employees actually build. Anything outside them is routed to a person, with the reason attached.

scheduled-digest

A periodic script whose output goes to Slack, email, sheets, or a webhook.

dashboard

Read-only data dashboards.

internal-web-app

Internal-only web apps and APIs. No payments.

everything else

Gets a reviewer and a clear next step. Nothing is silently dropped.

Privacy by architecture

Enabled, not exposed.
Metrics only. Consent always.

Your own report mirror: Every scan mirrors its raw report to a required S3-compatible bucket in your own cloud account, under your own credentials.
Dashboard upload is opt-in: The report is also written locally. Turn dashboard upload on, and an additional copy goes to the dashboard you configure.
Never executes: It reads traces. It does not run what it finds.

In plain terms

, in six lines.

The whole product, without the scroll.

01What it is: DKOD is a governed build pipeline for AI-built internal apps: an autonomous harness that lets organizations enable AI instead of blocking it.
02Discovery: dkod-signals is a single static Rust binary pushed via MDM (Jamf, Intune, Kandji) to every macOS, Windows, and Linux device. It runs once at low priority and inventories the apps employees built with Claude Code, Codex, Cursor, and other coding agents.
03The report: One privacy-safe, metrics-only JSON report per device. No source code, no file contents. Every report is mirrored to your own required bucket. An additional upload to your dashboard is opt-in.
04The picture: Device reports roll up into an org-wide view of AI-app risk posture: how many apps exist, how many are high-risk, and how many DKOD could rebuild safely.
05The rebuild: Qualifying apps are rebuilt from approved templates, scheduled digests, read-only dashboards, and internal web apps, through a governed pipeline in your own cloud, with deterministic gates, human approval, and a full audit trail.
06The control plane: Visibility and policy over the agents, skills, MCP servers, hooks, rules, permissions, models, budgets, secrets, and data access that AI coding work depends on.

FAQ

Questions,
answered.

Short answers. The docs have the long ones.

See what your org already built

Discovery with dkod-signals is available today. The governed build pipeline is now available.

Join early access Talk to us

Unblock AI for your whole org.Governed. Autonomous. Yours.

Nobody can see what AI already built. So nobody can say yes.

From shadow AIto a governed harness.

See

Set policy

Let it run

Agents do the work.Your policies set the rules.

See everything.Control it your way.

Agents

Skills

MCP servers

Hooks

Rules

Permissions

Models

Budgets

Secrets

Data access

Templates

Approvals and audit

Approved templates.Clear reasons.

Enabled, not exposed.Metrics only. Consent always.