AGISystem2 – Command Line Interfaces

AGISystem2 provides two distinct command-line interfaces, each designed for different use cases. Both interfaces share the same underlying engine and knowledge store, but differ in how users interact with them and what level of control they offer.

Two Modes, Different Purposes

When to Use Each Interface

Use the Raw CLI When:

• Automating tests – Write reproducible test scripts that verify reasoning behavior
• Building integrations – Script interactions for CI/CD pipelines or other tools
• Precise control – Need exact Sys2DSL commands with predictable results
• No external dependencies – Work offline without LLM API costs
• Auditing and compliance – Every command and result is deterministic and logged

Use the Chat Interface When:

• Exploring the system – Learn what AGISystem2 can do through conversation
• Quick demonstrations – Show capabilities without teaching syntax
• Building knowledge interactively – Teach facts in natural language
• Prototyping theories – Quickly test ideas before formalizing them
• Non-technical users – Allow domain experts to interact without coding

The Library Approach

Both command-line interfaces are convenience wrappers around the AGISystem2 JavaScript library. For production applications, most integrations will use the library directly through the APIs:

const AgentSystem2 = require('agisystem2');

// Create engine instance
const agent = new AgentSystem2({ profile: 'prod' });
const session = agent.createSession();

// Execute commands programmatically
session.run(['@f ASSERT Dog IS_A Animal']);
const result = session.run(['@q ASK "Is Dog an Animal?"']);
console.log(result.q.truth); // TRUE_CERTAIN

The library approach provides:

• Full programmatic control over all engine features
• Direct integration with application logic
• Custom session management and theory layering
• Event hooks for reasoning steps
• Memory-efficient batch operations

See the API Reference for complete documentation of the library interface.

Quick Start

# Start interactive session
./bin/AGISystem2Raw.sh

# Add facts
AGIS2> add Dog IS_A Animal
AGIS2> add Cat IS_A Animal

# Ask questions
AGIS2> ask Is Dog an Animal?
Result: TRUE_CERTAIN

# Batch mode
./bin/AGISystem2Raw.sh --exec "add Fire CAUSES Smoke" --json

# Start chat (requires LLM API key)
export OPENAI_API_KEY="sk-..."
./bin/AGISystem2.sh

# Natural conversation
You: Dogs are mammals that bark
AI: Got it! I've learned 2 fact(s):
    - Dog IS_A Mammal
    - Dog HAS_PROPERTY barks

You: Is a dog an animal?
AI: Yes, based on what I know...

Shared State

Both interfaces work with the same .AGISystem2/ directory structure in the current working directory:

.AGISystem2/
├── data/           # Concept storage, binary blobs
│   ├── concepts/   # Bounded diamond data
│   └── audit/      # Operation logs
└── theories/       # Theory files (.sys2dsl, .txt)
    ├── health_compliance.sys2dsl
    ├── law_minimal.sys2dsl
    └── scifi_magic.sys2dsl

Facts and theories created in one interface are immediately available in the other. This allows workflows like:

1. Use Chat interface to interactively build initial knowledge base
2. Export theory with /theories or save theory
3. Use Raw CLI for automated testing of that theory
4. Return to Chat for exploratory questions