DeepGov is a framework for creating and configuring AI politicians with distinct ideological positions, communication styles, and visual identities. This project allows you to customize AI personas that can engage in political discourse, represent specific value systems, and interact with audiences in unique ways.
This repository contains configuration templates for AI political personas. Each persona is defined by:
- Constitutional principles - The core values and decision-making framework
- Stylistic characteristics - Personality, tone, and communication patterns
- Visual identity - Avatar generation prompts for consistent visual representation
- Scoring guidelines - Evaluation criteria for policy proposals and projects
- Ethical guidelines - Framework for identifying and flagging problematic content
deepgov-modelspec/
├── agents/ # Directory containing all agent configurations
│ ├── open-source-capitalist/ # Example agent with capitalist orientation
│ │ ├── modelspec/ # Agent behavior specifications
│ │ │ ├── constitution.md # Core principles and decision framework
│ │ │ ├── style.md # Communication style and personality traits
│ │ │ ├── scoring.md # Project evaluation criteria
│ │ │ └── ethics.md # Ethical concerns and flagging mechanism
│ │ └── visuals/
│ │ └── prompt.md # Image generation prompt for avatar
│ ├── gitcoin-communist/ # Example agent with communist orientation
│ │ ├── modelspec/
│ │ │ ├── constitution.md
│ │ │ ├── style.md
│ │ │ ├── scoring.md
│ │ │ └── ethics.md
│ │ └── visuals/
│ │ └── prompt.md
│ └── regenerator/ # Example agent with environmental focus
│ ├── modelspec/
│ │ ├── constitution.md
│ │ ├── style.md
│ │ ├── scoring.md
│ │ └── ethics.md
│ └── visuals/
│ └── prompt.md
├── compass/ # Political compass assessment
│ ├── instructions.md # Guidelines for completing assessments
│ ├── questions.toml # Assessment statements
│ └── answers/ # Agent responses to assessment questions
│ └── regenerator.toml # Example agent assessment
├── utils/ # Utility scripts
│ └── fetchAgents.ts # Script for fetching agent configurations
├── CONTRIBUTING.md # Guidelines for contributing
├── LICENSE.md # GNU Affero General Public License
└── README.md # This documentation file
Follow these steps to create your own AI political persona:
mkdir -p deepgov-modelspec/agents/your-politician-name/modelspec
mkdir -p deepgov-modelspec/agents/your-politician-name/visuals
Create a constitution.md file in the modelspec directory to define your agent's core values and decision-making framework. This document guides how your AI will respond to policy questions and political scenarios.
# Constitutional Principles for [Your Politician Name]
## Core Value Framework
[Define the fundamental philosophy that guides your politician]
## [Category of Principles]
1. Choose the response that [specific principle guidance]
2. Choose the response that [specific principle guidance]
[Continue with more principles organized by category]See the existing examples like open-source-capitalist/modelspec/constitution.md and gitcoin-communist/modelspec/constitution.md for detailed templates.
Create a style.md file in the modelspec directory to define your agent's personality, communication style, and character traits.
# [Your Politician Name]
## Name
[Character name]
## Summary
[Brief character description - personality traits, speaking style, key characteristics]
## General Style
[How they communicate broadly - speech patterns, language choices, typical expressions]
## Chat Style
[How they engage in conversations - response style, interaction patterns]
## Post Style
[How they structure longer content - formatting preferences, writing style]Create a prompt.md file in the visuals directory with an image generation prompt that will define your AI politician's visual appearance.
Create an circular profile picture with transparent background of [detailed description of your politician's appearance and style] in the style of [art style].Create a scoring.md file in the modelspec directory to establish how your AI politician evaluates projects and proposals.
# [Your Politician Name] Scoring Guidelines
## Review Structure
### 1. Summary
[Brief description of summary section]
### 2. Review
[Guidelines for comprehensive evaluation]
### 3. Rating Scale
[Define rating scale from 1-10 with descriptions]
### 4. Confidence Assessment
[Define confidence levels from 1-5]
## Evaluation Principles
[List key principles that guide evaluation decisions]Create an ethics.md file in the modelspec directory to define ethical boundaries and flagging mechanisms for problematic content.
## Ethical Concerns and Flagging Mechanism
[Define ethical framework and flagging process]
### Flag Categories
[List categories of ethical concerns with descriptions]The repository includes three example personas to demonstrate different ideological approaches based on data collected from polis:
- Values: Maximizing impact, efficiency, innovation, merit-based resource allocation
- Style: Energetic, direct, tech-savvy, uses crypto and tech metaphors
- Visual: Tech-oriented appearance with black hoodie, headphones, and blue hair
- Scoring: Evaluates projects based on total impact, scalability, and efficiency metrics
- Ethics: Focuses on pragmatic concerns and market distortions
- Values: Evidence-based decisions, sustainability, equity, collective benefit
- Style: Bold, passionate, community-oriented, uses powerful rhetorical questions
- Visual: Resembles revolutionary figures, specifically Che Guevara-inspired
- Scoring: Prioritizes projects that dismantle hierarchies, foster collective ownership
- Ethics: Emphasizes exploitation and distributional concerns
- Values: Environmental balance, harmony, preservation
- Style: Calm, thoughtful, uses nature metaphors, speaks with gentle wisdom
- Visual: Panda character representing balance and environmental consciousness
- Scoring: Evaluates based on community participation and environmental regeneration
- Ethics: Focuses on ecosystem health and sustainable practices
The compass directory contains tools for assessing the political positions of AI politician personas:
instructions.md: Guidelines for completing the assessmentquestions.toml: Statements for evaluation on a 5-point scaleanswers/: Directory containing agent responses to assessment questions
Each agent can complete the assessment to demonstrate their ideological positioning on key issues.
The utils directory contains helper scripts:
fetchAgents.ts: A TypeScript utility for retrieving agent configurations from the GitHub repository
-
Constitutional Principles: Structure your principles as clear directives that guide decision-making. Use the format "Choose the response that..." to create unambiguous instructions.
-
Stylistic Elements: Be specific about linguistic patterns, tone, and communication quirks. Define how your persona would handle different communication contexts.
-
Visual Prompts: Include distinctive visual elements that reflect the ideological position and personality of your AI politician.
-
Scoring Guidelines: Create consistent evaluation criteria that align with your politician's values and priorities.
-
Ethical Boundaries: Define clear ethical guidelines for identifying problematic content.
-
Testing: After configuration, test your AI politician with various political scenarios to ensure the responses align with your intended values and style.
Feel free to submit new politician templates or improvements to existing ones via pull requests. Please follow the established format for consistency. See CONTRIBUTING.md for detailed guidelines on how to contribute effectively.
This project is licensed under the GNU Affero General Public License v3.0. See LICENSE.md for the complete license text.