Skip to main content

Basic Concepts

Understand the fundamental concepts behind Memfit AI.

Execution Modes

Instant Execution Mode

For clear, atomic tasks:

  • Entry: Direct to ReAct Loop
  • Best for: Single-step operations, queries, simple scans
  • Example: "What is the HTTP response from google.com?"

Plan-Execute Mode

For complex, multi-step tasks:

  • Entry: Through Coordinator
  • Best for: Audits, assessments, multi-phase operations
  • Example: "Perform a comprehensive penetration test"

The ReAct Loop

ReAct stands for Reasoning and Acting.

Cycle

Observation → Thought → Action → Observation → ...

Components

ComponentPurpose
ObservationGather information from environment
ThoughtReason about next steps
ActionExecute a tool or operation

Example

[Observation] User asked about port 443 on target.com

[Thought] I need to check if port 443 is open and identify the service.

[Action] PortScan(target: "target.com", ports: [443])

[Observation] Port 443 is open, running nginx/1.21.0

[Thought] Task complete. I'll report the finding.

Task Trees

For complex tasks, Memfit AI creates Task Trees.

Structure

Main Task
├── Subtask 1
│ ├── Sub-subtask 1.1
│ └── Sub-subtask 1.2
├── Subtask 2
└── Subtask 3

Properties

  • Dependencies: Tasks can depend on others
  • Parallelism: Independent tasks run in parallel
  • Recursion: Subtasks can spawn their own task trees

Human-in-the-Loop

Memfit AI involves you at critical decision points.

Plan Review

Before executing a plan:

  • Review proposed steps
  • Modify if needed
  • Approve to proceed

Action Confirmation

For sensitive operations:

  • File modifications
  • Network attacks
  • System changes

Clarification Requests

When information is missing:

  • Ask for target details
  • Request credentials
  • Confirm scope

Memory System

Memfit AI remembers and learns.

Short-term Memory

  • Current conversation context
  • Execution state
  • Recent observations

Long-term Memory

  • Previous interactions
  • Learned patterns
  • Error recovery strategies

Benefits

  • Improved accuracy over time
  • Contextual recommendations
  • Pattern recognition

Tools and Forges

Tools

Atomic capabilities:

  • File operations
  • Network requests
  • Code execution
  • Scanning

Forges

Scenario-specific packages:

  • Specialized prompts
  • Curated tool sets
  • Domain expertise
  • Workflow logic

Reliability Mechanisms

Self-Reflection

When things go wrong:

  • Analyze failure
  • Generate correction plan
  • Retry with improved context

Spin Detection

Prevents infinite loops:

  • Action repetition detection
  • Logic loop detection
  • Automatic intervention

Summary

ConceptDescription
ReAct LoopThought-Action-Observation cycle
Task TreeHierarchical task decomposition
CoordinatorCentral orchestration component
MemoryShort and long-term context
ToolsAtomic operations
ForgesScenario packages