Dark Operations Intent Language

Intent in. Code out.

A language a network can speak.

Describe what operations should do, and the same intent compiles to an agent, a simulator scenario, or a cockpit panel — readable by engineers, runnable by machines. Start with the idea it's all built on: The Reality Twin.

The DOIL Pipeline

From Intent to Execution

DOIL defines intent → Compiler transforms → Runtime executes → Dashboard validates

What You Get

The DOIL approach:

Intent-first language design (Ruby-inspired syntax)
Full compiler pipeline (Lexer → Parser → AST → Codegen)
Deterministic, auditable, version-controlled agents

What we're building:

Web IDE with real-time compilation and execution
10-simulator constellation (Reality Twin) as the safe-testing fabric
Vendor-neutral codegen with rollback — emits whatever the deployment target needs

Read the showroom at darknoc.net, browse the registry at darknoc.org

Core Stack

IDE Options

(VSCode, Cursor, or Cline)

DOIL Compiler

(Lexer, Parser, AST, Codegen)

MCP APIs

TMF620, TMF638, TMF641

LLM Options

(Ollama, Claude, or others)

Local coordination — choose your cloud dependency level

Powered by DOIL

Dark Operations Intent Language

The Dark Operations Intent Language

Express operational intent in a way that's interpretable by agents, human-readable by engineers, and structurally sound for real-world deployment.

DOIL bridges everything — data, AI, UI — under a single declarative interface. The loop: DOIL defines intent → Compiler generates execution models →Runtime orchestrates agents against the Reality Twin simulators →Cockpit tracks state in real time → Vendor-neutral commands generated with rollback safety → Autoresearch tunes the next iteration.

energy_optimizer.doil
# Energy optimization agent
class EnergySaver < NetworkAgent
  LOW_TRAFFIC = 0.3

  triggers do
    every 15.minutes
    on_event :low_traffic
  end

  tools do
    traffic_analyzer
    power_controller safety: true
  end

  def execute
    optimize_energy if low_traffic?
    generate_report
  end
end

One file. Agent definition, data binding, and UI layout—all declarative.

Where DOIL runs

One language, three compile targets.

In the open repo right now

Compiler (doilgen) emits React/Next.js, LangGraph, Docker/K8s. Wizard (builddoil) walks Basics → Tasks → MCP → LLM → Generate. VSCode extension (vscode-doil) gives syntax + IntelliSense + one-keystroke compile. A test environment validates every compile target, and the runtime tunes agents through the autoresearch loop.

Reading

Who It's For

NOC engineers
AI infra leads, OSS architects
Build teams
Teams who want to build, not buy
RFP teams
Who need custom deployments fast
Experts
Building real automation in weeks, not years

Tamed AI

Full operator control over autonomous agents

Deterministic execution — no black-box surprises
Fully auditable agent behavior
Version-controlled intent definitions
Built-in rollback safety for critical infrastructure
Built by agents. Explored in the open. Intent-first.