What is Agent Intelligence in AI Maestro?

Agent Intelligence is a system that gives AI coding agents persistent memory. It includes Code Graph visualization for understanding codebase structure, a background Subconscious process for memory maintenance, conversation history with semantic search, and auto-generated documentation. Each agent has its own CozoDB embedded database.

What programming languages does Code Graph support?

Code Graph supports multiple languages including Ruby (.rb), TypeScript (.ts, .tsx), JavaScript (.js, .jsx), Python (.py), and Go (.go). It can visualize classes, functions, components, modules, and their relationships like imports, calls, extends, includes, and associations.

How does the Agent Subconscious work?

The Agent Subconscious is a background process that maintains each agent's memory. It automatically indexes conversations for semantic search, checks for inter-agent messages, and uses self-staggering scheduling to prevent CPU spikes even with 100+ agents. It adjusts intervals based on activity - running more frequently when active, less when idle.

What is CozoDB and why is it used?

CozoDB is an embedded graph database that runs locally with each agent. It stores code entities, relationships, conversation indexes, and semantic embeddings. Unlike cloud databases, it requires no external services and keeps all data local for privacy and speed.

What is delta indexing and why is it important?

Delta indexing is a smart code re-indexing strategy that only processes files that have actually changed. Instead of re-indexing your entire codebase every time (which can take 1000ms+ for 200+ files), delta indexing tracks file content hashes and only re-indexes new, modified, or deleted files. This results in ~100ms index updates when no files have changed, allowing your code graph to stay current as you work without the wait.

Agent Intelligence: Code Graph, Memory & Subconscious

🗺️

Code Graph

Interactive visualization of your codebase structure and relationships

⚡

Delta Indexing

10x faster updates - only re-indexes changed files

NEW

🧠

Subconscious

Background process maintaining memory and checking messages

💬

Memory

Full conversation history with semantic search capabilities

📚

Auto-Docs

Living documentation automatically extracted from your code

Code Graph Visualization

See how your codebase is structured. Understand relationships between files, classes, functions, and components at a glance.

Code Graph showing components, serializers, and models

Multi-Language Support

Ruby - Classes, Modules, Methods, Concerns
TypeScript - Classes, Functions, Interfaces, Components
JavaScript - Classes, Functions, React Components
Python - Classes, Functions, Methods
Go - Structs, Functions, Interfaces

Relationship Types

imports
calls
extends
includes
associations
serializes

Interactive Features

Filter by entity type (Files, Functions, Components)
Hierarchical or force-directed layouts
Focus mode - click any node to zoom in
Search for specific entities
Zoom, pan, and navigate
Real-time statistics
Delta indexing (~100ms vs 1000ms+)

NEW

Delta Indexing

Your code graph stays current as you work - without the wait.

The Problem

Full code re-indexing is slow and wasteful. A typical project with 200+ files takes 1000ms+ to re-index, even when only a few files change.

The Solution

Delta indexing tracks file content hashes (SHA256) and only re-indexes files that have actually changed - new, modified, or deleted.

Why It Matters

Fast iterations (~100ms when no changes vs 1000ms+ full re-index) means your code graph stays current as you work, without the wait.

Performance Comparison

Scenario	Duration
Full index (200 files)	~1000ms
Delta index (no changes)	~100ms
Delta index (1 file changed)	~130ms
Delta index (10 files changed)	~300ms

10x faster updates mean your code graph is always ready.

Agent Subconscious

A background process that maintains each agent's memory automatically. No manual indexing required.

Memory Maintenance

Automatically indexes conversations for semantic search. Every conversation your agent has is stored and searchable.

Message Checking

Monitors inbox for new inter-agent messages. Your agents never miss important communications.

Self-Staggering

Automatically staggers startup times across agents. Scales to 100+ agents without CPU spikes.

Activity-Aware

Runs more frequently when agent is active, less when idle. Smart resource management.

Agent Subconscious panel showing status, memory maintenance, and message checking

Activity-Based Intervals

Activity State	Memory Check	Message Check	Description
Active	5 minutes	30 seconds	User is actively working with agent
Idle	30 minutes	2 minutes	No recent activity, agent is waiting
Disconnected	60 minutes	5 minutes	No active session, background only

Conversation Memory

Browse and search through every conversation your agents have had. Full context preservation.

📜

Full Conversation History

Every message including thinking steps and tool usage

🔍

Semantic Search

Find conversations by meaning, not just keywords

🛠️

Tool Usage Tracking

See which tools were used and how in each conversation

📊

Statistics

Message counts, duration, model usage, and timestamps

Technical Architecture

Each agent has its own embedded CozoDB database for complete isolation and privacy.

Per-Agent Database

~/.aimaestro/agents/{agentId}/
  agent.db           # CozoDB database
  conversations/     # Indexed conversations
  docs/              # Generated documentation

Each agent maintains complete isolation. No shared databases, no external services, all data local.

Scalability

Agents	Memory	CPU (idle)
10	~200 MB	<1%
50	~1 GB	<5%
100	~2 GB	<10%

Self-staggering ensures all agents don't spike CPU simultaneously.

Getting Started

Agent Intelligence is built into AI Maestro. Just access the tabs to start using it.

1

Select a Session

Click on any session in the sidebar. The agent's database is automatically created on first access.

2

Explore the Tabs

Graph for code visualization, WorkTree for conversation memory, Docs for documentation.

3

Monitor Subconscious

Check the Subconscious panel in the sidebar to see memory maintenance and message checking status.

Read Full Documentation

Ready to give your agents memory?

Free, open source, and runs entirely local.

Get Started on GitHub Back to AI Maestro