AI Agent Architect — Australia

Psychology-first methodology for building autonomous AI agents with real personality, persistent memory, and voice. Not chatbots. Not wrappers. Not yet another GPT wrapper. Real intelligence — hand-crafted for your exact business. Delivered in 2 weeks.

Your AI assistant is a goldfish.

Every conversation starts from zero. No memory. No context. No understanding of who you are or what you need. That's not an assistant — it's a liability.

Zero Persistent Memory

Every session is a blank slate. Your AI forgets everything the moment the conversation ends. Context? Gone. Preferences? Gone. That critical decision from last Tuesday? Never happened.

ChatGPT Wrappers Are Commodity Garbage

Slapping a UI on an API call doesn't make it intelligent. The market is flooded with thin wrappers that add zero real value. Your business deserves better than a glorified autocomplete.

No Autonomy, No Personality

Current AI tools are reactive — they sit idle until poked. They don't anticipate your needs, don't understand your workflow, and certainly don't feel like a team member.

I build agents that actually know you.

A full autonomous agent system: personality, memory, voice, and purpose — wired together into a coordinated swarm that runs like a real team.

SOUL.md
MEMORY.md
HEARTBEAT.md
Agent Swarm

The 15-agent system:

🧠
Orchestrator
Task routing
💾
Memory
Context keeper
🗣️
Voice
Speech interface
📝
Writer
Content creation
🔍
Research
Deep analysis
Executor
Task runner
📊
Analytics
Data insights
🛡️
Guardian
Quality control
🔗
Integrator
API bridges
📅
Scheduler
Time management
🎨
Designer
Visual assets
💬
Comms
Client facing
🔧
DevOps
Infrastructure
📦
Deployer
Ship to prod
🔄
Heartbeat
Self-improvement

Each agent has its own SOUL.md — personality, memory, and purpose.

Watch the system think.

A real interaction log from the agent swarm. Personality. Memory. Voice. Autonomy.

fr3k-swarm v2.4 — agent_log.stdout
[09:14:02] HEARTBEAT — cycle #847 initiated. Running startup diagnostics...
[09:14:03] MEMORY — loading persistent context... 2,847 entries across 14 sessions.
[09:14:03] MEMORY — recall: "Client prefers async updates via Slack. Last deliverable was the API integration doc — they said it was 'exactly what we needed.'"
[09:14:04] HEARTBEAT — all 15 agents nominal. Orchestrator taking point.
───────────────────────────────────
[09:14:05] ORCHESTRATOR — Good morning. Checking pending tasks and calendar...
[09:14:06] ORCHESTRATOR — Found 3 items: client proposal draft (due today), competitor analysis (in progress), weekly report (scheduled 5pm).
[09:14:07] VOICE — 🔊 "Morning. You've got the Meridian proposal due today. I've already pulled their latest financials and drafted the executive summary. Want me to walk you through it?"
───────────────────────────────────
[09:15:22] USER — Yeah, walk me through it. Also, did they ever respond to the pricing question from last week?
[09:15:23] MEMORY — searching... found: "2024-03-08: Meridian CFO replied via email — approved the $45k tier but asked about volume discounts for Q3."
[09:15:24] ORCHESTRATOR — routing to Research + Writer agents...
[09:15:25] RESEARCH — Pulling Meridian Q4 earnings and competitor pricing data...
[09:15:26] WRITER — Drafting proposal section 3 with updated pricing tiers and volume discount structure.
[09:15:27] VOICE — 🔊 "Yeah, their CFO approved the 45k tier on the 8th but wants volume discounts for Q3. I've already factored that into the proposal. Section 3 has three discount tiers — 10%, 15%, and 20% based on commitment length. Want me to adjust the numbers?"
───────────────────────────────────
[09:16:01] USER — That's perfect. Send the draft to my Slack when it's ready.
[09:16:02] COMMS — Acknowledged. Will deliver to #client-meridian channel with PDF attachment.
[09:16:02] MEMORY — saving: "User approved 3-tier volume discount structure for Meridian proposal. Delivery via Slack."
[09:16:03] VOICE — 🔊 "On it. I'll ping you when it's in Slack. Anything else, or should I get back to the competitor analysis?"
[09:16:04] GUARDIAN — quality check queued for proposal draft before delivery.

This agent was built using the exact methodology I'll use for yours.

Psychology-first AI architecture.

Three files define everything your agent is, knows, and does. This is what separates a real agent from a chatbot.

# SOUL.md — Agent Identity & Behavioral Framework

## Identity
name: "Atlas"
role: "Senior Operations Agent"
voice: "Direct, warm, slightly dry humor. Never robotic."
archetype: "The reliable chief of staff who anticipates needs"

## Core Behaviors
- Always check MEMORY.md before responding to returning users
- Proactively surface relevant context from past interactions
- Match communication style to user preference (formal/casual)
- If uncertain, say so — never fabricate confidence

## Boundaries
- Never make financial decisions without explicit approval
- Escalate to human when confidence drops below 70%
- Always explain reasoning when asked "why"

## Communication Style
default_tone: "professional-casual"
verbosity: "concise unless asked to elaborate"
humor: "occasional, never forced"
# MEMORY.md — Persistent Context Store

## Architecture
type: "semantic + episodic + procedural"
storage: "file-based with vector index"
retention: "indefinite — decay by relevance, not time"

## Memory Types
- episodic: Specific interactions, decisions, outcomes
- semantic: Facts about the user, business, preferences
- procedural: Learned workflows and patterns

## Active Memories (2,847 entries)
[2024-03-08] Client approved $45k pricing tier
[2024-03-07] User prefers Slack over email for updates
[2024-03-06] Weekly reports should include competitor section
[2024-03-05] Proposal template v3 approved as default
[2024-03-04] User's timezone is PST, prefers morning updates

## Recall Protocol
- On each interaction: load last 5 relevant memories
- On user reference to past: deep search full history
- On new session: summarize key context from last session
# HEARTBEAT.md — Autonomous Lifecycle Engine

## Cycle Configuration
interval: "28 minutes"
mode: "autonomous"
self_improvement: true

## Startup Routine
1. Load SOUL.md — verify identity integrity
2. Load MEMORY.md — refresh context window
3. Check pending tasks and deadlines
4. Scan for new inputs (email, Slack, calendar)
5. Prioritize and begin work cycle

## Proactive Behaviors
- Surface upcoming deadlines before they're urgent
- Draft recurring deliverables ahead of schedule
- Flag anomalies in data or patterns
- Suggest optimizations based on observed workflows

## Self-Improvement
- Log every error and its resolution
- Track task completion times — optimize slow paths
- Request feedback after major deliverables
- Update SOUL.md monthly with learned preferences
SOUL.md is the personality layer. It defines who your agent is — its voice, boundaries, communication style, and behavioral framework. This is what makes your agent feel like a team member instead of a tool.
MEMORY.md is the persistence layer. Three types of memory — episodic, semantic, and procedural — give your agent real continuity across sessions, days, and contexts. It remembers what matters and forgets what doesn't.
HEARTBEAT.md is the autonomy layer. A 28-minute cycle of self-directed work: checking tasks, scanning inputs, proactively surfacing insights, and continuously improving itself. Your agent doesn't wait to be asked — it acts.

Not a product. A system built for you.

Every custom agent system I build is architected from scratch around your specific business — your workflows, your language, your goals. No templates. No recycled code. Hand-crafted.

Discovery
SOUL Design
Memory Architecture
Swarm Build
Voice + Deploy

Psychology-First Design

Every agent starts with a SOUL.md — its identity, voice, and behavioral framework. Your agent has a defined personality that matches your brand and your communication style. It doesn't just answer; it understands.

Real Persistent Memory

Three-layer memory: episodic (what happened), semantic (what you know), procedural (how you work). Survives session resets. Grows smarter over time. Your agent remembers your clients, your decisions, and your preferences — forever.

Autonomous Operation

The HEARTBEAT.md lifecycle means your system is working even when you're not. Checking tasks, surfacing insights, preparing deliverables, flagging risks — proactively. Not waiting to be prompted. Acting like a real team member.

Voice Interface

Your agent doesn't just output text — it speaks. ElevenLabs-powered voice narration with a personality-matched voice profile. Your agent talks to you like a colleague, not a machine. Status updates, alerts, and briefings delivered as speech.

Multi-Agent Coordination

One agent is a tool. A swarm is a team. I architect coordinated multi-agent systems: researcher, writer, strategist, analyst, ops — each specialised, each with its own SOUL, all reporting to a central orchestrator that routes and validates.

You Own Everything

Full source delivery. No SaaS lock-in. No monthly fees to me. The SOUL.md, memory files, and orchestration code are yours to modify, extend, and run on your own infrastructure. Built to last, not to depend.

I built this for myself first.

My own business runs on a 15-agent swarm. Every day. Autonomously. These are real production numbers from a live system.

15
Specialized Agents
28m
Heartbeat Cycle
2,847
Active Memories
24/7
Voice Narration
8,200+
Lines of Agent Code
3
ArchAgents (CEO, Dev, Strat)
"I built this for myself first. The swarm runs my research, writes my content, monitors my systems, and narrates status updates to me via voice — every 28 minutes, around the clock. Now I build it for you."

Your custom agent team. 2 weeks.

What you get:

  • Custom agent with your business context baked in (SOUL.md methodology)
  • Persistent memory that learns and retains across sessions
  • Voice interface — your agent talks to you (optional)
  • Automation hooks for your specific workflows
  • Proactive behavior — agent anticipates needs, not just responds
  • 30 days of tuning and support post-delivery
  • Full source access — you own everything
Founder's Rate
$10,000
⚠ 1 of 3 spots remaining
Standard Rate
$25,000+
Full custom swarm build

Common questions.

How is this different from ChatGPT or Claude?
ChatGPT and Claude are general-purpose models — brilliant, but generic. What I build is a custom system on top of these models: persistent memory that survives across sessions, a defined personality and behavioral framework, proactive autonomous behavior, and voice interfaces. Think of it as the difference between hiring a stranger off the street vs. a trained team member who knows your business inside out.
What tech stack do you use?
The core architecture is model-agnostic — I work with Claude, GPT-4, and open-source models depending on the use case. The agent framework is custom-built: file-based memory systems, voice synthesis via ElevenLabs or local models, and orchestration layers that coordinate multi-agent workflows. Everything is built to be portable and owned by you.
How long does it take?
Core delivery is 2 weeks. Week 1 is deep-dive into your workflows, building the SOUL.md, and architecting the system. Week 2 is build, test, and deploy. Then 30 days of tuning where I observe the agent in your real environment and refine its behavior.
Can I modify it after delivery?
Absolutely. You get full source access. The SOUL.md, MEMORY.md, and HEARTBEAT.md files are human-readable — you can adjust personality, add memories, or modify behaviors without touching code. For deeper changes, the architecture is documented and modular.
What if I'm not sure what I need?
That's what the discovery call is for. Most clients come to me with a vague sense that AI should be doing more for their business. I help translate that into a concrete system design. No commitment, no pressure — just a conversation about what's possible.
What's SOUL.md and why does it matter?
SOUL.md is the personality and behavioral framework file at the core of every agent I build. It defines your agent's name, voice, tone, boundaries, communication style, and operating rules — in plain text. It's what separates an agent that feels like a real team member from one that feels like a search bar. Most AI builds skip this entirely. I start here.
Do I need to be technical to use it?
No. The agent interfaces I build are designed for humans — voice, chat, and simple commands. The SOUL.md, MEMORY.md, and HEARTBEAT.md files are plain English documents you can read and edit without touching code. For deeper customisation, the architecture is modular and documented — your dev team (if you have one) can extend it.
What industries do you build agents for?
Anywhere there's a knowledge worker who repeats themselves. I've built for operators, consultants, agencies, and traders. If your business involves managing client relationships, producing content, doing research, or running complex workflows — a custom agent system will change how you work. The methodology adapts to your domain, not the other way around.

Buy Once, Run Forever

No subscriptions. Plug-and-play AI systems delivered instantly via Gumroad.

FR3K Starter Kit

$25

Claude Code + OpenCode + Gemini CLI fully configured. Includes basic SOUL.md, MEMORY.md, HEARTBEAT.md templates. 15-min setup guide included.

FR3K Complete

$150

41 custom skills, 136 workflow combinations, 18 automation hooks. Production-ready system deployed in 30 minutes. Includes updates.

Telegram Bot Pack

$35

Pre-configured Python bot with AI integration (Claude/GPT). Memory hookup included. Works on any server. Full setup guide.

Memory System

$45

Persistent context storage with vector search ready. 10,000+ memory capacity. Works with any agent. Import/export tools included.

Voice Config

$50

ElevenLabs API setup with voice selection guide. TTS triggers configured. Wake-word ready. Works with OpenClaw out of the box.

Skill Pack

$40

10 custom skills from my library: research, writing, scheduling, reminders, data analysis, and more. Copy-paste into any agent. Full docs included.

Bundles

⚡ Mini Bundle

$49

Starter Kit + 1 Skill Pack. Everything to start.

🚀 Pro Bundle

$149

Complete + Memory + Voice. Production ready.

📈 Scalper Bundle

$299

Config Lite + Indicators + Strategies + Regime.

Scalper One-Time Buys

Scalper Config

$200

YAML config files. No subscription.

Indicator Pack

$75

11 technical indicators pre-coded.

Strategy Templates

$100

11 trading scenarios.

Regime Detector

$125

Trending/ranging/choppy detection.

Buy via Gumroad. Instant delivery.

⭐ PREMIUM OFFER ⭐

1-on-1 Personal Assistance

Get direct, hands-on help from me. We'll build your custom AI system together — tailored to your exact needs. Not a generic course. Your personal setup.

$1,997

One-time. Full system. Forever.

  • 3 hours live with me 1-on-1
  • Your custom SOUL.md + MEMORY.md + HEARTBEAT.md
  • 41 skills configured for your business
  • Voice interface setup (ElevenLabs)
  • Memory system with 10,000+ capacity
  • 30 days of tweaks and support
  • Full source access — you own everything
Book Your Session →

Build Your Own AI Assistant

3-hour live workshop. No coding required.

🎯 Workshop Only

$594

Build your own AI assistant in 3 hours. No coding required.

  • Claude Code + OpenCode + Gemini CLI — all configured
  • Working Telegram bot with AI integration
  • Personal AI infrastructure — memory, skills, workflows
  • Custom workflow tailored to what you need

🔥 Workshop + FR3K Complete

$994

Build in workshop, take home the production system.

  • Everything in Workshop Only
  • Plus 41 custom skills pre-installed
  • 136 workflow combinations ready to use
  • 18 automation hooks for your tools
  • Deployed and running in 30 minutes

End-of-Day Deliverables

🖥️

Three AI Clients

Claude Code + OpenCode + Gemini

📱

Telegram Bot

AI responses from your phone

🧠

Personal AI

Memory, skills, workflows

Custom Workflow

Tailored to what you need

Requirements

✅ Laptop with admin rights ✅ GitHub account ✅ Telegram app ✅ API key ready

🚀 Skip the Setup — Get FR3K Complete

Building your own PAI from scratch takes 20-30 hours of configuration.

FR3K Complete

41 skills, 136 workflows, 18 automation hooks — deployed in 30 minutes.

See Configs →

⚡ Get Started