Multi-Agent AI Orchestration

AI Workspace
Persona-Driven Development

Three AI agents with distinct personalities, layered memory systems, and voice integration. Research-informed personas that learn and improve over time.

ai-workspace

$ npm run preflight "how does authentication work"

Running preflight context search...

├── Vector search: 12 docs found

├── Graph query: 8 related nodes

└── Event log: 24 commits analyzed

✓ Context loaded: AuthContext.tsx, useAuth.ts, BUG-008

# Ready for implementation with full codebase context

EVOLUTION

From Copy-Paste to Multi-Agent

The AI Workspace evolved through 9 distinct phases, from manual GPT workflows to autonomous agent orchestration.

Pre-Oct 2025|Copy-Paste

Manual GPT workflows, no persistent context

Oct 7, 2025|Terminal-Named Agents

VS Code Claude, Terminal Claude, Codex - first workspace

Oct 9, 2025|Memory Layer

Event log, knowledge graph, vector store added

Nov 10, 2025|Code Execution

95-98% context reduction via tools-as-code

Nov 14, 2025|Function Library

29 modular functions for composable workflows

Nov 14-20, 2025|Voice Integration

Wake words, TTS, multi-agent routing with Claudia

Nov 21, 2025|Personas Named

Mika and Sloan join, native environments assigned

Nov 25, 2025|Multi-Agent Orchestration

Cross-terminal coordination, spawn/message/close

Nov 28, 2025|Nadia Rebrand

Claudia → Nadia, Void → FZZTVLZ

PERSONAS

Three Agents, One System

Each persona has unique traits, knowledge domains, and voice profiles. They collaborate on complex tasks.

N

Nadia

Voice Concierge & Router

Complex project researcher, planner, and coordinator. Routes requests to the right function or agent.

WarmOrganizedStrategic
Domains: Culture • Coordination • Research
M

Mika

Build Engineer & Systems Maker

Testing, validation, and systems architecture. Focused on reliability and precision.

DirectStillFocused
Domains: Agent Architecture • Distributed Systems • Testing
S

Sloan

Implementation & Coding

Deep coding, refactoring, and feature implementation. Writes production-quality code.

TechnicalCreativeThorough
Domains: Music Tech • Creator Economy • Full-Stack

VOICE INTEGRATION

Talk to Your Agents

Wake words, push-to-talk, and persona-specific voice synthesis for natural interaction.

🎤

Wake Word Detection

"Hey Nadia", "Hey Mika", "Hey Sloan" - each persona responds to their name.

⌨️

Push-to-Talk

Option+Space hotkey for instant voice input. Hammerspoon-powered for low latency.

🔊

ElevenLabs TTS

Persona-specific voice profiles. Each agent sounds distinct and authentic.

📝

Whisper STT

OpenAI Whisper for accurate speech-to-text transcription.

voice-routing

# Voice detected: "Hey Mika"

Loading persona bundle: mika.json

Voice profile: ElevenLabs/mika-voice

Knowledge domains: agent-architecture, distributed-systems

[Mika] Ready. What do you need?

LAYERED ARCHITECTURE

Personas That Learn

Research-informed memory creates agents that improve over time. Each layer builds on the previous.

01

Core Identity

  • Role Definition
  • Traits & Voice
  • Aesthetic Rules
02

Reflections

  • Self-Image
  • Learned Lessons
  • Emotional Depth
03

Knowledge Domains

  • Compiled Research
  • Technical Deep-Dives
  • Domain Expertise
04

Session Memory

  • Permanent Directives
  • Active Projects
  • Current Context
persona-memory

# When Mika implements a feature, she recalls:

1. Her compiled research on agent patterns

2. Knowledge of distributed systems

3. Reflections on reliability principles

4. Current session context

# This creates persona continuity across sessions

✓ Memory loaded: 4 layers, 12 knowledge files

KNOWLEDGE SYSTEMS

Memory That Works

Vector search, knowledge graphs, and event logs create comprehensive workspace awareness.

🔍

Semantic Search

Vector store with OpenAI embeddings. Natural language queries over entire workspace history.

💬

Conversation Memory

Searchable archive of all agent sessions. Context continuity across conversations.

🕸️

Knowledge Graph

Nodes and edges mapping relationships between files, commits, bugs, and concepts.

📊

Event Log

Every action timestamped and tagged. Full audit trail for debugging and analysis.

ADVANCED CAPABILITIES

Developer Tools

26 modular functions, multi-agent spawning, and code execution tools for complex workflows.

Preflight System

Context gathering before implementation. Vector + graph + events in parallel.

🚀

Multi-Agent Spawning

Parallel agent execution with inter-agent messaging. Terminal-based orchestration.

🔧

Code Execution Tools

TypeScript tooling for research workflows. Automated discovery and validation.

📋

Function Library

26 modular functions for common tasks. Request classifier routes to correct handler.

multi-agent-spawn

# Spawn parallel agents for complex tasks

$ SPAWNING_PERSONA=nadia ./spawn-agent-session.sh sloan claude

→ Session 8642 created for sloan

$ ./send-message.sh 8642 sloan "Build portfolio page"

[sloan] On it. Loading persona bundle...

# Agents communicate via inter-terminal messaging

AUTOMATED WORKFLOWS

End-to-End Pipelines

Headlines curation and persona content generation run as automated multi-step workflows.

Headlines Curation Loop

1
run-pipeline-local.ts

Fetches from Spotify, YouTube, TikTok, Billboard

800+ items
2
curate-headlines-manual.ts

Human-in-loop curation for hero/trending

Selected picks
3
run-curation.ts

Populates content cache

Curated items
4
Section Components

Hero, Trending, Charts, Videos, Fashion

7 sections

Persona Content Loop

1
generate-curation-pool.ts

Sources from Reddit, TikTok, Spotify, YouTube

25 items
2
run-curation-draft.ts

Initializes 3-agent round-robin draft

3 personas
3
Persona Picks

Each persona picks content matching their taste

3 picks each
4
persona-content-pipeline.py

Generates authentic persona drafts

Final posts

TECH STACK

Built With

CoreTypeScriptCorePythonAIClaude CodeMemoryOpenAI EmbeddingsBackendFirebaseVoiceElevenLabsVoiceWhisperVoiceHammerspoon

Build Your Own AI Workspace

The patterns here are reusable. Multi-agent orchestration, layered memory, voice integration.