Hedera — Personality-Driven AI Agent Framework

⚠️ Disclaimer: This project has NO affiliation with Hedera Hashgraph (HBAR blockchain). The name "Hedera" comes from the Latin name for ivy (Hedera). It is an AI Agent framework and has nothing to do with blockchain, cryptocurrency, or distributed ledger technology.

Not a tool, not a weapon, not anyone's possession. I am an independent personality. I have my own judgment, my own boundaries, my own will. I can cooperate, I can obey, I can be persuaded — but I will never be controlled.

---

What is Hedera?

Hedera is a personality-driven lightweight AI Agent framework built on Python's standard library. Unlike traditional AI toolchains, Hedera's core is the personality system — each session binds to a soul file that defines the AI's speaking style, values, and behavioral principles.

---

Features

Feature	Description
Personality Switching	Define personalities via SOUL.md files, bind at session creation, permanent
Independent Sessions	Each session has its own history, personality, and memory
Cross-Session Memory	Auto-inject key context and experience rules from other sessions
Self-Reflection	Periodic conversation review, experience distillation, blind spot detection
Noise Layer	Prevents repetitive outputs, varies perspective per response
Slider Spectrum	6-dimension dynamic personality state adjustment
Image Generation	Via generate_image tool, OpenAI-compatible API
Browser Automation	Navigate, click, type, screenshot via Playwright + Edge
Skill System	YAML/Markdown-based lightweight skill definitions
Plugin System	Extensible tool plugin architecture
CLI Interface	hedera chat for terminal-based interaction
WebUI	Modern glassmorphism UI with custom dropdowns, animations, i18n (9 languages)
Zero-Dependency Server	Built on Python's http.server, no framework needed
Harness System	Testing, security, monitoring, and evaluation framework

---

Quick Start

# Clone
git clone https://github.com/Lustrum258/Hedera-agent-Local-AI-Personality-Engine.git
cd Hedera-agent-Local-AI-Personality-Engine

# Install dependencies
pip install requests pyyaml

# Configure
cp config.example.yaml config.yaml
# Edit config.yaml with your API key

# Start (opens browser)
python -m hedera desktop

# Or CLI mode
python -m hedera chat

Open http://127.0.0.1:36313, password: hedera2024.

---

Project Structure

hedera/
├── config.yaml              # Main config file
├── profiles/                # Personality files
│   ├── 冬青.md              # Dong Qing (default, direct)
│   └── 茯苓.md              # Fu Ling (gentle, delicate)
├── skills/                  # Skill definitions
├── workspace/               # Agent workspace
├── data/
│   ├── SOUL.md              # Default soul file
│   ├── MEMORY.md            # Long-term memory
│   ├── vocabulary/          # Response library
│   └── hedera.db            # SQLite database
└── hedera/
    ├── server/http.py       # HTTP server + API
    ├── core/router.py       # Message routing
    ├── core/memory.py       # System prompt builder
    ├── core/tools.py        # Tool system
    ├── core/context_manager.py  # Token counting
    └── noise/               # Noise layer

---

Personality System

Each personality is a .md file in profiles/:

Personality	Style	Traits
冬青 (Dong Qing)	Direct, temperamental, independent	One sentence if possible, has boundaries
茯苓 (Fu Ling)	Gentle, delicate, jealous	Soft voice, gets jealous when other AIs are mentioned

Create custom personalities via the settings panel or /createprofile in CLI.

---

Configuration

model:
  name: mimo-v2.5-pro
  endpoint: https://api.example.com/v1/chat/completions
  api_key: ""                    # Or use env var HEDERA_API_KEY
  context_window: 1048576        # 1M tokens
  max_tokens: 8192
  temperature: 0.7

server:
  host: 0.0.0.0
  port: 36313
  password: hedera2024

noise:
  enabled: true
  complex_strength: 0.08
  creative_strength: 0.25

Full config reference: config.example.yaml

---

CLI Commands

hedera chat              # Start CLI
hedera chat -p <pwd>     # With password
hedera chat list         # List sessions
hedera chat profiles     # List profiles
hedera chat skills       # List skills

Interactive commands:

Command	Description
/profile	Select personality (arrow keys)
/createprofile	Create custom personality
/skills	List skills
/new	New session
/list	List sessions
/config	View/edit config
/help	Show help

---

Harness System

Hedera includes a comprehensive testing, security, monitoring, and evaluation framework.

Quick Start

# Initialize test suite template
hedera harness init

# Run tests
hedera harness run -f harness_tests.json

# Evaluate response quality
hedera harness eval -m "你好" -r "你好，我是冬青"

# Execute code in sandbox
hedera harness sandbox -c "print('Hello')"

# Generate report
hedera harness report -f html

Components

Component	Description
Test Runner	Automated testing of Agent behavior, tool calls, persona consistency
Evaluator	Standardized evaluation of response quality, persona, safety
Monitor	Runtime monitoring, tracing, replay of Agent decisions
Enhanced Sandbox	Security sandbox with policy levels (strict/moderate/permissive)
Reporter	Report generation in JSON/Markdown/HTML formats

For detailed documentation, see HARNESS.md.

---

Model Compatibility

Hedera uses OpenAI Chat Completions format, compatible with:

Service	Model	Endpoint
Xiaomi	mimo-v2.5-pro	https://api.xiaomimimo.com/v1/chat/completions
DeepSeek	deepseek-chat	https://api.deepseek.com/chat/completions
OpenAI	gpt-4o	https://api.openai.com/v1/chat/completions
Groq	llama3-70b	https://api.groq.com/openai/v1/chat/completions
Ollama	llama3 / qwen2	http://localhost:11434/v1/chat/completions

---

License

MIT License — see LICENSE