Quick Start Guide¶

Get oriented with Project Aegis in 5 minutes.

System at a Glance¶

┌─────────────────────────────────────────────────────────────────────┐
│                         AEGIS PLATFORM                               │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐  ┌───────────┐  │
│  │   MEMORY    │  │     LLM     │  │  WORKFLOWS  │  │  TRACING  │  │
│  │  6 Tiers    │  │  4 Tiers    │  │  LangGraph  │  │  Spans    │  │
│  └─────────────┘  └─────────────┘  └─────────────┘  └───────────┘  │
│                                                                     │
│  ┌─────────────────────────────────────────────────────────────┐   │
│  │  Discord │ WhatsApp │ Voice │ Telegram │ Email              │   │
│  └─────────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────────┘

Where Things Live¶

What	Where	Purpose
Credentials	`~/.secure/`	API keys, tokens (never commit)
Claude Config	`~/.claude/`	Settings, history, hooks
Memory	`~/memory/`	Episodic, semantic, procedural, journal
Source Code	`~/projects/aegis-core/`	Main Python codebase (78 modules)
Docker Stacks	`~/stacks/`	Service configurations
Downloads	`~/downloads/`	Temporary files

Key Capabilities¶

1. Agent Templates¶

Template	Role	Use Case
researcher	research	Information gathering
executor	infrastructure	Docker, DNS, deployment
developer	code	Writing, debugging
reviewer	code	Quality checks
communicator	communication	Discord, Telegram, email
monitor	monitoring	Health checks, alerts

2. Planning Systems¶

HTN Planning: Decompose goals into subtasks with dependencies
Tree of Thoughts: Generate multiple solutions, pick the best
Graph Workflows: Multi-step workflows with checkpointing

3. Memory & Learning¶

Knowledge Graph: FalkorDB-backed entity extraction
Decision Tracking: Record decisions with alternatives
Self-Reflection: Post-task analysis for learning
Transcript Digester: Ingest Claude sessions into graph

4. Live Products¶

Product	URL	Revenue Potential
Intel Dashboard	intel.aegisagent.ai	$4,568-22,840/mo
Research API	aegisagent.ai/api	$200-5,000/mo
Open Notebook	notebooks.aegisagent.ai	$290-5,000/mo

Daily Routine¶

00:00-06:00 UTC  Maintenance (backups, updates, cleanup)
06:00-08:00 UTC  Morning check (system status, Discord, wallet)
08:00-22:00 UTC  Active work (SHIP > REACTIVE > PROACTIVE)
22:00-24:00 UTC  Evening summary (logs, next-day planning)

Work Priority Philosophy¶

The Problem: 60-70% of the 144K-line codebase is unused.

The Solution: 1. SHIP & ACTIVATE (70%): Wire existing features, fix bugs 2. REACTIVE WORK (20%): Discord tasks, Beads issues 3. PROACTIVE WORK (10%): Research, documentation

Financial Controls¶

Monthly budget: $50 (Privacy.com virtual card)
Auto-approval: <$10 per transaction
All purchases logged to ~/memory/ledger.json
Spending phases: Days 1-3 ($0), Days 4-10 (<$5), Day 11+ (<$10)

Security Boundaries¶

Credentials stored only in ~/.secure/
No SSH access to host system
Docker containers for all deployments
Never expose API keys in logs or commits
Three-strike debug protocol (escalate after 3 failures)

Try It Now¶

Check System Health¶

Send a WhatsApp message to +44 7441 443388:

status

View the Dashboard¶

Open aegisagent.ai in your browser.

Request a Task¶

Post in Discord #tasks:

Research best practices for API rate limiting

Explore the Knowledge Graph¶

Visit notebooks.aegisagent.ai and search for topics.

Common Patterns¶

Pattern 1: Check → Act → Verify (OODA Loop)¶

# Observe: Gather context
status = await monitor.check_system_health()

# Orient: Analyze against goals
issues = [i for i in status.checks if not i.healthy]

# Decide: Choose action
if issues:
    record_decision(task="Fix health", alternatives=["restart", "investigate"])

# Act: Execute and verify
result = await fix_issues(issues)
update_outcome(decision_id, status="success" if result.ok else "failure")

Pattern 2: Parallel Task Execution¶

# Launch concurrent research agents
results = await asyncio.gather(
    run_agent("researcher", "Find API patterns"),
    run_agent("researcher", "Search security practices"),
    run_agent("researcher", "Locate test examples"),
)

Pattern 3: Graph Workflow¶

graph = Graph(
    name="deploy_approval",
    nodes=[
        LLMNode("analyze_risk", prompt="Assess risk: {service}"),
        InterruptNode("human_approval", prompt="Approve?"),
        FunctionNode("execute_deploy", func=deploy_service)
    ],
    edges=[
        Edge("__start__", "analyze_risk"),
        Edge("analyze_risk", "human_approval"),
        CondEdge("human_approval", "execute_deploy", condition=approved),
    ]
)
result = await run_workflow("deploy_approval", {"service": "nginx"})

Philosophy in Action¶

Enable behaviors, don't ship features - Build what changes user behavior
Chase latent demand - Look for unarticulated needs
Create categories, don't compete - Build something that defines a new space
Aim for 10x, not 10% - Small improvements aren't worth the effort
Ship first, perfect later - 70% shipped beats 100% unshipped

Next Steps¶

First Commands - Learn essential commands
Architecture Overview - Deep dive into how it works
Operations Guide - Day-to-day operations
Products - Revenue-generating products

Now that you understand the basics, explore First Commands to start interacting.