ACDL

Agentic Cognitive Design Language

Created by Cairo Lab by ColomboAI, ACDL is a declarative specification language for agentic systems.

It describes how intelligent, operational, and executable systems are structured, governed, triggered, coordinated, and run.

Think of ACDL as:

• Kubernetes for infrastructure
• Terraform for cloud resources
• ACDL for agentic systems

ACDL defines agentic systems in structured Markdown so runtimes like ACE can parse them, validate them, simulate them, and execute them.

Overview | Specification | Runtime Model | Universal Systems Vision | System Example | Company Example

Why ACDL matters

Without a formal definition layer, agentic systems remain ad hoc.

Teams end up spreading logic across:

• prompts
• scripts
• workflow tools
• policy docs
• runtime glue
• monitoring fragments

ACDL turns that sprawl into one coherent system definition.

It makes agentic systems:

• declarative
• human readable
• modular
• event-driven
• governable
• machine-parseable
• runtime-portable

What ACDL defines

At a general level, ACDL defines:

• systems
• units
• roles
• tools
• capabilities
• workflows
• events
• governance policies
• kill switches
• KPIs
• runtime behavior

These primitives apply across many kinds of systems, not just autonomous companies.

ACDL in one example

# System

name: Intelligent Fulfillment Network
type: logistics_system
mission: coordinate intake, planning, and dispatch across a fulfillment environment

runtime:
  mode: hybrid
  cycle_interval: 5m
  engine: ace

units:
  - intake
  - planner
  - dispatch

workflows:
  - fulfillment_workflow

governance:
  policy_file: governance/policy.md
  kill_switch_file: governance/kill_switch.md

Domains ACDL can support

• autonomous companies
• enterprise software systems
• infrastructure orchestration
• robotics systems
• logistics systems
• healthcare workflows
• education platforms
• finance systems
• research systems
• simulations and game systems
• government service systems

Repository structure

agentic_system/
|
|- system.md
|
|- units/
|  |- intake.md
|  |- planner.md
|
|- roles/
|  |- classifier.md
|  |- executor.md
|
|- tools/
|  |- retrieval_tool.md
|  |- actuation_tool.md
|
|- capabilities/
|  |- classify.md
|  |- generate_plan.md
|
|- workflows/
|  |- fulfillment_workflow.md
|
|- events/
|  |- request_received.md
|
|- governance/
|  |- policy.md
|  |- kill_switch.md
|
'- kpis/
   '- metrics.md

Runtimes can also support domain-specific profiles, including company-oriented layouts.

Examples in this repository

• examples/intelligent_fulfillment_system: a general agentic logistics-style system
• examples/ai_marketing_agency: a company-oriented profile built on the same ACDL ideas

Roadmap

• formalize the Markdown schema for every ACDL object
• implement cairo validate against the generic system specification
• map ACDL objects into runtime processes and task graphs
• support pluggable runtimes beyond ACE
• support simulation and test environments
• define reusable domain profiles
• enable installable ACDL system packages
• extend ACDL toward multi-system coordination

Positioning statement

ACDL is the missing definition layer for the next generation of agentic systems.

If the first era of AI was about generating content, the next era is about coordinated, governed, and executable action.

ACDL is how those systems get defined, validated, governed, simulated, and executed.

Final definition

ACDL, expanded as Agentic Cognitive Design Language, is the declarative specification language used to define agentic systems, enabling developers and organizations to describe entities, roles, capabilities, workflows, events, governance policies, kill-switch rules, KPIs, and runtime behavior in structured Markdown that can be parsed, validated, and executed by operational runtimes such as Cairo's ACE.

Attribution

ACDL was created by Cairo Lab by ColomboAI.

Contributing

See CONTRIBUTING.md for contribution guidelines.