Therapist

AI Psychological Therapist — CLI Edition

A pure JavaScript + Node.js CLI application that provides empathetic, professional psychological support through a multi-Agent pipeline. Designed with a focus on user experience, it actively probes context, detects psychological defense mechanisms, and delivers tailored therapeutic responses.

---

Features

• Multi-Agent Pipeline — Three specialized agents work in sequence: Organizer → Strategist → Therapist
• 6 Psychological Dimensions — Covers emotion, psychology, personality, attachment, self-worth, and defense mechanisms
• Adaptive Strategy — Dynamically adjusts approach based on detected user defense and engagement
• Session Persistence — Full conversation history stored locally for multi-session continuity
• Multi-LLM Support — Pluggable provider architecture (Volcano, DeepSeek, OpenAI, Zhipu, Baidu, MiniMax, Ali, Tiwen)
• Context Engineering — 6-layer prompt architecture from base persona to therapeutic output
• Ethics & Boundaries — Built-in safeguards: no diagnosis, emergency referral for crisis, privacy protection

---

Architecture

3-Agent Pipeline

User Input
    │
    ▼
┌─────────────┐
│  Organizer  │  ← Extracts known/missing knowledge across emotion / psychology / personality
└─────────────┘
    │
    ▼
┌─────────────┐
│ Strategist  │  ← Detects defense mechanisms, plans conversation strategy
└─────────────┘
    │
    ▼
┌─────────────┐
│ Therapist   │  ← Synthesizes final empathetic response with targeted questions
└─────────────┘
    │
    ▼
User Response

6-Layer Context Engineering Stack

Level	Layer	Description
1	Base Persona	Fixed therapist identity, ethical boundaries, professional background
2	Professional Glossary	~90 psychological terms auto-injected when detected
3	Dynamic Boundary Control	Rules: no diagnosis, crisis referral, privacy protection
4	Knowledge (Organizer Output)	Known/missing knowledge across 3 dimensions + session delta
5	Strategy (Strategist Output)	Conversation thread + defense analysis + strategy selection
6	Therapeutic Output	Final user-facing response

---

Project Structure

src/
├── entry/                    # CLI entry, argument parsing, REPL loop
│   ├── index.js             # Main entry point
│   ├── setup.js             # LLM provider configuration commands
│   └── session.js           # Session management
├── dispatch/                 # Core dispatcher orchestrating the 3-Agent pipeline
│   └── dispatcher.js
├── agents/                   # Three core agents
│   ├── organizer/           # Organizer Agent — knowledge extraction
│   ├── strategist/          # Strategist Agent — defense detection & strategy planning
│   └── therapist/           # Therapist Agent — final response synthesis
├── models/                   # Data models
│   ├── state.js             # Global session state
│   ├── knowledge.js         # Knowledge structures (emotion/psychology/personality)
│   └── strategy.js          # Strategy output structure
├── utils/                    # Utilities
│   ├── llm.js               # LLM HTTP API caller (multi-endpoint, multi-format)
│   ├── prompt.js            # Prompt builders for each agent
│   ├── history.js           # Context window management
│   ├── storage.js           # Session file persistence (~/.therapist/sessions/)
│   └── ui.js                # CLI rendering (banners, bubbles, colors, tables)
└── config/                   # Configuration
    ├── llm.js               # Reads llm.json, exports active LLM config
    ├── constants.js         # System prompts, boundary rules, glossary
    └── questionBank.js      # 6-dimension psychological assessment question bank

---

Quick Start

# Install dependencies
npm install

# Run (single prompt mode)
npm start

# Run (interactive / continuous chat mode)
npm run start:interactive
# or
node src/entry/index.js --interactive

---

CLI Commands

Command	Description
(any text)	Chat with the therapist
help / ?	Show command help
clear / cls	Clear the screen
q / quit / exit	Exit the program
setup	Interactive LLM provider configuration
setup list	Show current LLM configuration
setup use <provider>	Switch LLM provider
setup test	Test LLM connection
session new	Create a new session
session list	List all saved sessions
session select <id>	Switch to a session
session delete <id>	Delete a session
session info <id>	Show session details
session clear	Clear current session data

---

LLM Configuration

Configure your API credentials in llm.json. The active provider is set via the provider field.

# Interactive setup
node src/entry/index.js
> setup

# Or edit llm.json directly

Supported providers: volcano (default), deepseek, openai, zhipu, baidu, minimax, tiwen, ali

---

Psychological Question Bank

The Strategist Agent triggers assessment questions across 6 dimensions. Questions are camouflaged as casual conversation (penetration-type) or confirm previously inferred conclusions (verification-type):

• emotion — short-term state ("I feel...")
• psychology — medium-term stress ("lately I...")
• personality — long-term traits ("since I was a child...")
• attachment — relationship patterns
• selfWorth — self-esteem
• defense — how the user protects themselves

---

Extensibility

The architecture is designed for future expansion:

• Port/Server Mode — Replace the CLI entry with an HTTP server; dispatcher.js logic is reusable
• Web Frontend — A basic Express server scaffold exists in /web
• Multi-User Support — Auth layer (email + SMTP verification) + SQLite database planned

---

License

MIT