Core Technical Capabilities
The core philosophy
Traditional
Prompt → Model → Output
CAIRO
Software → Workflow → Memory → Local Intelligence → MC-1 → Frontier
Result
Better Systems Create Better Intelligence
Getting Started
From zero to autonomous operation in five steps.
Step 1 — Launch Cairo
Access the Cairo platform through ColomboAI.
Step 2 — Set Your First Goal
Enter a goal — Cairo plans, executes, verifies, and optimizes.
Step 3 — Deploy Skills & Plugins
Install modular capabilities or create your own.
Step 4 — Launch a Swarm
Use Long-Horizon Swarm Intelligence for complex, multi-week projects.
Step 5 — Deploy an Autonomous Company
Run 24/7 AI-operated organizations at scale.
Installation
Cairo runs on all major platforms and deployment environments.
Desktop
- Windows
- macOS
- Linux
- Download Desktop
Local Runtime
- Docker
- Source Build
- Enterprise Deployment
- Enterprise Deployment
Core Concepts
CAIRO is built around a 9-stage operational loop that continuously improves itself.
Operational Consciousness Loop
Operational Intelligence
AI systems that monitor, analyze, decide, and act in real time across operational environments — not just respond to prompts.
Operational Consciousness Infrastructure
The full stack that makes autonomous, self-healing, self-improving operations possible at scale.
Cairo for Everyone
Cairo is designed for all users, not just developers.
Individual Users
Anyone can deploy AI assistants and operational tools directly from the Cairo platform.
- Research assistants
- Productivity assistants
- Financial planning tools
- Learning assistants
- Personal automation systems
Developers
Developers can build advanced AI systems using Cairo Core.
- AI agents
- Multi-agent systems
- Automation frameworks
- Operational AI systems
- Autonomous AI companies
Organizations
Enterprises can deploy Cairo operators to manage operational workflows continuously.
- Enterprise automation
- Financial operations
- Marketing operations
- Supply chain management
- Infrastructure monitoring
The Cairo Capability Hierarchy
Cairo systems are structured in layers of increasing capability.
Skills
Atomic AI agents
Skills are atomic AI agents capable of performing specific tasks. Skills represent the smallest unit of operational intelligence.
Plugins
Domain-specific orchestrations
Plugins are orchestrations of multiple skills designed to solve domain-specific problems.
Marketing Plugin
- Content creation
- Campaign management
- Analytics tracking
Legal Plugin
- Contract analysis
- Legal research
- Regulatory monitoring
Finance Plugin
- Financial analysis
- Reporting
- Risk monitoring
AI Companies
Highest level of operational systems
AI companies combine multiple plugins and skills coordinated by Cairo operators, running 24/7 autonomously.
Cairo Architecture
A layered intelligence system — each layer enables different capabilities.
Brain Layer
Reasoning and planning powered by ColomboAI-MC-1.
- Multi-step reasoning
- Decision planning
- Tool orchestration
- Operational strategy generation
Perception Layer
Detects capabilities across interfaces, APIs, and environments.
- Tools
- APIs
- System capabilities
- Application interfaces
Orchestration Layer
Coordinates multi-agent workflows and task execution.
- Multi-agent coordination
- Parallel task execution
- Operational workflow planning
- Execution monitoring
Tool Layer
Executes actions across software systems, APIs, and infrastructure.
- APIs
- Applications
- Enterprise systems
- Web services
- Automation frameworks
Memory Layer
Stores operational knowledge and learns from past outcomes.
- Learning from past tasks
- Remembering operational context
- Improving performance over time
Observability Layer
Monitors system performance and execution traces.
- Execution tracing
- Debugging
- Performance monitoring
- System evaluation
Governance Layer
Ensures safe and compliant autonomous operations.
- Permission management
- Risk controls
- Compliance monitoring
- Operational policy enforcement
Software-First Intelligence
Distribute intelligence across software — not just models
Traditional AI relies on large models for every task. CAIRO distributes intelligence across software, workflows, memory, orchestration, connectivity, swarm coordination, and models — routing only what is necessary to each layer. This dramatically reduces cost while improving reliability.
Conscious Efficiency Engine
Routes each task to the cheapest capable layer.
Routing Policies
Per-task rules for escalation, fallback, and cost caps.
Context Compression
Compresses operational context before model calls to minimise token spend.
Hybrid Inference Architecture
Four-layer inference stack from native code to frontier models
CAIRO's inference stack escalates requests through four layers, using the cheapest sufficient resource for each task. Most operations never reach Layer 3 — they are resolved by software or local models.
Deterministic code paths — zero LLM cost.
Nemo 3 Nano Omni, Gemma 4, Qwen 3.6 Local — private, fast, offline-capable.
Mid-tier reasoning for complex operational tasks.
Frontier intelligence for highest-complexity decisions.
Long-Horizon Swarm Intelligence
Chain Swarm Intelligence Protocol (CSIP)
Most AI systems fail on long projects and large codebases because context degrades. CAIRO solves this with the Chain Swarm Intelligence Protocol — a coordinated system of specialised swarms that hand off work, regenerate when entropy is detected, and maintain continuity through Operational DNA™.
Planning Swarm
Decomposes long-horizon goals into executable task graphs.
Research Swarm
Gathers, synthesises, and validates domain knowledge.
Coding Swarm
Implements, tests, and reviews code at scale.
Validation Swarm
Verifies correctness, coverage, and quality.
Recovery Swarm
Detects failures and regenerates broken execution paths.
Security Swarm
Audits for vulnerabilities and enforces policy.
Optimization Swarm
Reduces cost, latency, and entropy over time.
Swarm Regeneration™
When entropy, hallucination, or context degradation is detected, CAIRO generates a fresh swarm to continue the task — preserving strategic continuity without losing progress.
Operational DNA™
A structured representation of an operation's identity, goals, history, and strategy that is inherited by child swarms and evolved over time.
Long-Horizon Task Kernel (LHTK)
Checkpoints, continuity recovery, compression, and entropy monitoring that preserve execution state across multi-week operations.
Memory System
Engineered Memory Protocol (EMP)
Traditional AI memory forgets everything between sessions. CAIRO's EMP provides four persistent memory layers with compression, graph-based knowledge storage, and cross-session continuity.
Working Memory
Active context for the current task.
Strategic Memory
Mid-term summaries and operational patterns.
Long-Term Operational Memory
Persistent knowledge graphs across sessions.
Historical Memory
Archived execution graphs and outcomes.
Memory Graph stores
Operational Document Intelligence (ODIL)
Software-first document processing for operational knowledge
ODIL converts documents into operational knowledge using MarkItDown integration and software-first extraction pipelines — without burning tokens on raw document parsing.
Supported Formats
Extraction Outputs
- Operational Knowledge
- Workflow Extraction
- Entity Extraction
- Operational Graph Generation
Operational Connectivity
DCRS — Dynamic Capability Retrieval System + Nango Integration
CAIRO discovers and routes to capabilities dynamically through the DCRS, and connects to SaaS systems via Nango — handling OAuth, sync jobs, webhooks, and event streams automatically.
DCRS Capabilities
- Capability Discovery
- Operational Routing
- Dynamic Tool Generation
- Capability Graphs
Nango OAuth Connectivity
- OAuth SaaS Connectivity
- Enterprise Connectivity
- Sync Jobs
- Webhooks
- Event Streams
Supported Systems
ACE — Autonomous Company Engine
Cairo's runtime for executing autonomous operational organizations
ACE orchestrates AI operators, departments, workflows, operational queues, governance systems, and safety controls — converting declarative ACDL definitions into running AI organizations.
ACE Operational Hierarchy
Continuous Operations
- Monitoring marketing campaigns
- Processing payments
- Analyzing financial data
- Responding to customer requests
- Managing enterprise workflows
Governance & Safety
- Operational policy enforcement
- Approval mechanisms
- Risk monitoring
- Cost controls
- Kill switches
Kill switches allow operators or administrators to pause or stop autonomous systems when necessary.
ACDL — Autonomous Company Definition Language
Declarative language for defining autonomous operational systems
ACDL is the declarative language used to define autonomous operational systems in Cairo. These definitions are compiled by ACE into running operational systems.
Typical ACDL Structure
ACDL Resource Types
Company
Defines the organization and operational structure.
- Mission
- Departments
- Governance policies
- Revenue model
Departments
Group operators responsible for specific domains.
- Marketing
- Finance
- Legal
- Operations
Operators
AI workers responsible for executing tasks. Can run continuously or respond to events.
- Campaign Operator
- Billing Operator
- Research Operator
- Support Operator
Plugins
Orchestrate multiple skills into domain functionality.
- Marketing Plugin
- Accounting Plugin
- Legal Plugin
Skills
Perform individual tasks.
- Generate Ads
- Send Invoice
- Analyze Data
- Draft Contract
Workflows
Define sequences of operational steps triggered by events.
- payment_received → verify_payment → launch_campaign
Events
Trigger automation across operational systems via EIL.
- payment_received
- invoice_failed
- customer_signup
- contract_request
Policies
Enforce governance rules across operations.
- High transaction approval rules
- Operational safety constraints
Kill Switch
Emergency control over autonomous systems.
- Pause operator
- Pause department
- Pause company
- Safe-degrade operations
- Hard stop system
KPIs
Measure operational performance. KPI signals can trigger corrective workflows.
- Monthly revenue
- Campaign ROI
- Conversion rate
- Customer acquisition cost
Event Intelligence Layer (EIL)
Autonomous reaction to events across digital environments
EIL enables CAIRO to react automatically to events from any connected system — routing them to operators, workflows, or departments without human intervention.
Event Processing Pipeline
EIL Features
- Event Architecture
- Event Sources (any connected system)
- Event Routing
- Event-Driven Execution
- Autonomous Triggers
- Operational Awareness
Meta-Harness Intelligence
Route tasks to the right execution harness
CAIRO's Meta-Harness layer maintains a registry of specialised execution harnesses and routes tasks to the best harness for each job — from browser automation to software engineering to custom enterprise integrations.
Hermes
General-purpose agentic harness for operational workflows.
Codex
Software engineering harness — code, test, deploy.
Claude Code
Claude-native agentic coding integration.
OpenClaw
Browser automation harness for web environments.
Browser Agents
Visual web navigation and form execution.
Custom Harnesses
Build your own harness and register it to the registry.
Cairo Software Engineer
Autonomous software engineering through specialised swarms
Cairo Software Engineer is a full autonomous engineering system — capable of understanding large codebases, architecting solutions, writing and testing code, and deploying with self-healing recovery loops.
Engineering Swarm Agents
Self-Healing Engineering
- Build Failure Recovery
- Test Failure Recovery
- Deployment Failure Recovery
- Automatic Recovery Loops
Autonomous Companies
24/7 AI-operated organizations
CAIRO enables fully autonomous companies — organizations where AI operators manage operational workflows continuously, coordinated by a shared company memory, goals, and governance system.
Sales Company
Continuous pipeline generation, outreach, and deal management.
Marketing Company
Campaign creation, publishing, analytics, and optimisation.
Engineering Company
Software development, testing, deployment, and maintenance.
Support Company
24/7 customer support across all channels.
Research Company
Continuous market, competitive, and domain research.
Custom Company
Define your own departments, operators, and workflows.
Company Infrastructure
Plugins
Domain-specific orchestrations of skills
Plugins bundle multiple skills into cohesive operational capabilities. Developers can create plugins, publish them to the Marketplace, and monetise through subscriptions or usage fees.
Plugin Architecture
- Skill composition
- Domain-specific orchestration
- Governance policies
Plugin Lifecycle
- Create → Test → Publish → Marketplace → Deploy
Skills
Atomic AI capabilities — the smallest unit of operational intelligence
Skills are the atomic building blocks of CAIRO. A skill performs a single task. Skills compose into Plugins, which compose into Autonomous Companies.
Skill Structure
- Input schema
- Execution logic
- Output schema
- Error handling
Skill Composition
- Skills → Plugins
- Plugins → Companies
- Shared skill marketplace
Skill Examples
- Research
- Data analysis
- Content generation
- Code writing
- Contract analysis
- API operations
MCPs
Model Context Protocol connectors
MCPs are protocol connectors that extend CAIRO's capability reach to external systems, APIs, and data sources using the Model Context Protocol standard.
MCP Architecture
- Connector Design
- Authentication
- Permissions
- MCP Marketplace
MCP Builder
- Define connector schema
- Map to external API
- Publish to Marketplace
- Version management
Security & Governance
Multi-layer safety for autonomous operations
Autonomous systems require robust safety mechanisms. CAIRO provides governance at every layer — from individual prompt filtering to full sovereign deployment controls.
Prompt Firewall
Filters malicious or policy-violating inputs before execution.
Operator Controls
Per-operator permissions, rate limits, and scope restrictions.
Enterprise Controls
SSO, audit logs, data residency, and role-based access.
Sovereign Controls
Air-gapped deployment and full data sovereignty options.
Kill Switches
Pause operator, department, company, or hard-stop the system.
Risk Engine
Real-time risk scoring and approval gates for high-risk actions.
CAIROSOUL.MD
The Constitution of CAIRO
CAIROSOUL.MD is the foundational constitution that defines CAIRO's identity, values, and operating principles. Every CAIRO system inherits these doctrines.
Open Source
Contribute to CAIRO
CAIRO is open to contributors. The codebase, swarm benchmarks, and documentation standards are all open for community participation.
Contributing
- Development Setup
- Architecture Overview
- Coding Standards
- Pull Requests
- Documentation Standards
Benchmarks
- Swarm Benchmarks
- Long-Horizon Benchmarks
- Memory Benchmarks
- Cost Efficiency Benchmarks
Roadmap
What's coming to CAIRO
Current Development
- Swarm Regeneration™ v2
- Operational DNA™ inheritance
- LHTK checkpointing
- ODIL graph generation
Upcoming Features
- MC-2 multimodal model
- Physical AI integration
- Advanced ACDL compiler
- Marketplace monetization
Research Areas
- Long-horizon continuity
- Swarm coordination protocols
- Operational cost optimization
- Self-improvement loops
Future Protocols
- CSIP v2
- Cross-company orchestration
- Sovereign AI deployment
- Physical world operators
Community
Join the CAIRO ecosystem