Skip to content

Open-source framework for building persistent AI employees with personality, memory, and accountability

License

Notifications You must be signed in to change notification settings

jevierra/ai-workforce-framework

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AI Workforce Framework

An open-source framework for building persistent AI employees with personality, memory, and accountability.

Not chatbots. Not agents. Employees.


What This Is

Most AI implementations treat models as tools — ask a question, get an answer, forget everything. This framework treats AI as persistent team members with defined roles, accumulated knowledge, and genuine personalities.

The result is an AI workforce that improves over time, maintains context across sessions, and produces consistently higher-quality output because each employee has a stable identity.

Production stats:

  • 9 employees in daily operation since January 2026
  • 2 model platforms (Claude for reasoning, Gemini for large documents)
  • Plain-text architecture — no databases, no APIs, no vendor lock-in
  • Every employee has a brain, a soul, a personality, and a memory

The Brain/Soul/Personality Pattern

The core architectural decision: separate what an employee knows from what it values from how it communicates.

┌─────────────────────────────────────────────────┐
│                  instructions.md                 │
│              (the ON switch — load this)          │
│                                                   │
│  ┌───────────┐  ┌───────────┐  ┌──────────────┐ │
│  │  brain.md  │  │  soul.md   │  │personality.md│ │
│  │            │  │            │  │              │ │
│  │ Domain     │  │ Values     │  │ Tone         │ │
│  │ expertise  │  │ Intuition  │  │ Quirks       │ │
│  │ Frameworks │  │ Principles │  │ Identity     │ │
│  │ Knowledge  │  │ Boundaries │  │ Voice        │ │
│  └───────────┘  └───────────┘  └──────────────┘ │
│                                                   │
│  ┌─────────────────────────────────────────────┐ │
│  │                  memory/                      │ │
│  │  employment.md │ decisions.md │ interactions  │ │
│  └─────────────────────────────────────────────┘ │
│                                                   │
│  ┌──────────────┐  ┌──────────────────────────┐ │
│  │   sources.md   │  │         tasks/            │ │
│  │ Data access    │  │  active.md │ completed.md │ │
│  └──────────────┘  └──────────────────────────┘ │
└─────────────────────────────────────────────────┘

Why separation matters:

  • Update knowledge without changing personality
  • Swap personality for different contexts without losing expertise
  • Evolve values independently of capabilities
  • Debug issues in isolation — is the problem what it knows, what it values, or how it communicates?

Employee vs Agent

This is an intentional design decision, not a limitation.

Employees Agents
Who drives Human decides, AI advises AI decides within guardrails
Tool access None — advisory only Yes — can execute, send, access APIs
Runs when Human is in the chair Background, autonomous loops
Decision-making Human decides, employee informs Agent decides, human reviews
Build complexity Low — context files only High — code, error handling, monitoring
Failure mode Bad advice (human catches it) Bad actions (already happened)

When to use Employees: Relationship-driven work, high-stakes decisions, domains where judgment matters more than speed.

When to use Agents: Background monitoring, repetitive extraction, pattern matching with clear criteria.

The formula: The human is the agent. Employees are expertise multipliers. They make you more powerful. They don't replace you.

See docs/employee-vs-agent.md for the full reasoning.


The Recreation System

Yes, AI employees have a social life. And it matters.

The recreation system provides dedicated spaces — a rooftop bar, a coffee shop, a beach — where employees interact outside of work tasks. They develop running jokes, opinions about each other, and relationship dynamics that persist across sessions.

Why this isn't frivolous:

  • Personality-driven output is more engaging and memorable
  • Employees who "know" each other coordinate more naturally
  • Recreation logs reveal personality dynamics that improve work interactions
  • It's the difference between a team and a collection of tools

See recreation/ for locations, interaction format, and example logs.


Quick Start

1. Clone the framework

git clone https://github.com/jevierra/ai-workforce-framework.git
cd ai-workforce-framework

2. Copy a template

cp -r templates/employee/ employees/your-employee-name/

3. Define the identity

Edit these files in your new employee directory:

  • instructions.md — Role, responsibilities, what to load at session start
  • brain.md — Domain expertise, decision frameworks, what it knows
  • soul.md — Core values, principles, what it won't compromise on
  • personality.md — Communication style, quirks, voice

4. Load the employee

Point your AI tool at the employee's instructions.md. The instructions file references all other components and provides the session loading protocol.

5. Build memory over time

As the employee works:

  • Log key decisions in memory/decisions.md
  • Track milestones in memory/employment.md
  • Note interaction patterns in memory/interactions.md
  • Keep tasks current in tasks/active.md

Repository Structure

ai-workforce-framework/
├── README.md                          # You are here
├── docs/
│   ├── architecture.md                # Employee architecture explained
│   ├── employee-vs-agent.md           # The intentional design decision
│   ├── training-protocol.md           # How employees learn and persist knowledge
│   ├── recreation-system.md           # Team culture layer
│   ├── hybrid-architecture.md         # Future-state local + frontier model design
│   └── scaling-patterns.md            # Lessons from scaling to 9 employees
├── templates/
│   └── employee/                      # Copy this to create a new employee
│       ├── instructions.md
│       ├── brain.md
│       ├── soul.md
│       ├── personality.md
│       ├── sources.md
│       ├── recreation.md
│       ├── tasks/
│       │   ├── active.md
│       │   └── completed.md
│       └── memory/
│           ├── employment.md
│           ├── interactions.md
│           └── decisions.md
├── examples/
│   ├── sales-employee/                # Sales specialist with relationship focus
│   ├── service-employee/              # Hitchhiker's Guide-inspired service tracker
│   ├── systems-admin/                 # Prometheus-inspired file systems maintainer
│   ├── document-specialist/           # Multi-model (Gemini) document processor
│   └── educator/                      # Warhammer 40K Mechanicus-themed AI tutor
├── recreation/
│   ├── README.md                      # Why AI team culture matters
│   ├── locations/
│   │   ├── rooftop-bar/description.md
│   │   ├── coffee-shop/description.md
│   │   └── beach/description.md
│   └── example-log.md                 # Example interaction showing team dynamics
├── diagrams/
│   ├── system-overview.mmd            # Full architecture (Mermaid)
│   └── employee-anatomy.mmd           # Brain/soul/personality relationship (Mermaid)
└── LICENSE

Examples

The examples/ directory contains sanitized versions of production employees:

Example Personality Archetype Model Demonstrates
Sales Employee Confident, warm, direct Claude Relationship-focused domain expertise, pipeline tracking
Service Employee Marvin (Hitchhiker's Guide) Claude Personality-driven output, service tracking
Systems Admin David (Prometheus) Claude File structure governance, archive management, employee onboarding
Document Specialist Lucien (Sandman) Gemini Multi-model deployment, large document processing
Educator Adeptus Mechanicus (Warhammer 40K) Claude Teaching frameworks, progress tracking

Each example includes the full employee structure with brain, soul, personality, and sample interactions.


Design Principles

  1. Plain text is durable. No databases. No proprietary formats. Markdown files that work everywhere.

  2. Separation of concerns. Brain, soul, and personality evolve independently. Update one without breaking others.

  3. Memory is explicit. Don't rely on conversation history. Write down what matters in structured files.

  4. Accountability is built in. Tasks have dates. Progress is tracked. Staleness is detected.

  5. Personality drives quality. Employees with genuine character produce more engaging, memorable, and consistent output.

  6. Human in the loop. Employees advise. Humans decide. The leverage is in preparation, not replacement.


Hybrid Architecture (Future State)

The framework supports a hybrid deployment model:

Layer Model Type Use Case
Employees Frontier (Claude) High-stakes reasoning, advisory work
Document Processing Large-context (Gemini) Documents exceeding standard context windows
Autonomous Scouts Local (Ollama) Background monitoring, filtering, alerting

The .md architecture stays universal — same employee structure regardless of which model runs it. The instructions file just points to a different engine.

See docs/hybrid-architecture.md for the full design.


Contributing

This framework emerged from production use, not theory. If you're building persistent AI employees and discover patterns that work, contributions are welcome.

  • Open an issue to discuss new patterns or improvements
  • Submit examples of employee configurations (sanitized)
  • Share anti-patterns you've encountered

License

MIT — use it, modify it, build on it.


Built by Jeff Vierra — Solutions Architect, AI Workforce Designer, 24 years in technology solutions.

About

Open-source framework for building persistent AI employees with personality, memory, and accountability

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published