ux change, prompt fix, readme02.md, .yaml, .toml

README01.md

@@ -0,0 +1,391 @@
+
+# CGJI01 — Multi-Agent Jungian Intelligence System
+
+CGJI01 is an interactive multi-agent psychological reasoning system designed to simulate deep, multi-perspective inquiry rooted in analytical psychology and related disciplines.
+
+At its core, the system begins with a Jungian analytical framework and expands into a network of specialized agents (Dream, Shadow, Myth, Epistemic, Buddhist Psychology, etc.), enabling layered interpretation and synthesis of human inquiry.
+
+---
+
+## 🧠 SYSTEM OVERVIEW
+
+### Flow of Interaction
+
+1. User initiates conversation ("Fire")
+2. Input is routed to **Agent Jung (default entry point)**
+3. Jung Agent:
+   - Interprets query
+   - Retrieves internal knowledge (RAG)
+   - Optionally performs web search
+   - Produces structured response
+4. System continues in **Jung follow-up loop**
+5. User may:
+   - Continue dialogue (Fuel)
+   - Select another agent (via dynamic choice system — upcoming)
+6. Selected agent takes control (agent switching)
+7. Multi-agent perspectives accumulate
+8. Final synthesis (planned via Synthesizer Agent)
+
+---
+
+## 🧩 ARCHITECTURE
+
+### Backend (FastAPI)
+
+api_server.py
+↓
+conversation_engine.py
+↓
+agent (jung_agent.py)
+↓
+LLM (Groq / LLaMA 3.1)
+
+---
+
+### Core Components
+
+#### 1. Conversation State (`conversation_manager.py`)
+
+Maintains:
+
+- Global history
+- Per-agent histories
+- Current active agent
+- Stage tracking
+- Initial query (anchor for agent switching)
+
+```python
+state = {
+    history: [],
+    agent_histories: {},
+    current_agent: "jung",
+    initial_query: "...",
+    stage: "jung"
+}
+
+2. Conversation Engine (conversation_engine.py)
+
+Central router controlling flow.
+
+Responsibilities:
+
+Handles session lifecycle
+Injects web-search context
+Locks active agent
+Routes to correct agent
+Stores memory
+Returns structured response to UI
+
+3. Agent Layer
+
+Each agent is modular:
+
+agents/
+    jung_agent.py
+    dream_agent.py
+    shadow_agent.py
+    ...
+
+🧠 AGENT JUNG — CORE DESIGN
+
+Agent Jung is the primary intelligence layer and philosophical anchor of the system.
+
+🎯 Objective
+
+To simulate a Jungian analytical psychologist that:
+
+Interprets user input symbolically
+Encourages introspection
+Guides toward individuation
+Avoids shallow informational responses
+Maintains psychological depth
+⚙️ Functional Pipeline
+1. Input Extraction
+user_input = state.history[-1]["content"]
+2. Web Search (Conditional)
+
+Triggered via heuristic:
+
+should_use_web(query)
+
+Adds:
+
+state.web_context
+3. RAG (Retrieval-Augmented Generation)
+
+From:
+
+/rag/jung/jung.index
+/rag/jung/jung_texts.txt
+
+Using:
+
+FAISS
+sentence-transformers (MiniLM)
+retrieve(query, k=3)
+4. Context Constructio# CGJI01 — Multi-Agent Jungian Intelligence System
+
+CGJI01 is an interactive multi-agent psychological reasoning system designed to simulate deep, multi-perspective inquiry rooted in analytical psychology and related disciplines.
+
+At its core, the system begins with a Jungian analytical framework and expands into a network of specialized agents (Dream, Shadow, Myth, Epistemic, Buddhist Psychology, etc.), enabling layered interpretation and synthesis of human inquiry.
+
+---
+
+## 🧠 SYSTEM OVERVIEW
+
+### Flow of Interaction
+
+1. User initiates conversation ("Fire")
+2. Input is routed to **Agent Jung (default entry point)**
+3. Jung Agent:
+   - Interprets query
+   - Retrieves internal knowledge (RAG)
+   - Optionally performs web search
+   - Produces structured response
+4. System continues in **Jung follow-up loop**
+5. User may:
+   - Continue dialogue (Fuel)
+   - Select another agent (via dynamic choice system — upcoming)
+6. Selected agent takes control (agent switching)
+7. Multi-agent perspectives accumulate
+8. Final synthesis (planned via Synthesizer Agent)
+
+---
+
+## 🧩 ARCHITECTURE
+
+### Backend (FastAPI)
+
+api_server.py
+↓
+conversation_engine.py
+↓
+agent (jung_agent.py)
+↓
+LLM (Groq / LLaMA 3.1)
+
+---
+
+### Core Components
+
+#### 1. Conversation State (`conversation_manager.py`)
+
+Maintains:
+
+- Global history
+- Per-agent histories
+- Current active agent
+- Stage tracking
+- Initial query (anchor for agent switching)
+
+```python
+state = {
+    history: [],
+    agent_histories: {},
+    current_agent: "jung",
+    initial_query: "...",
+    stage: "jung"
+}
+2. Conversation Engine (conversation_engine.py)
+
+Central router controlling flow.
+
+Responsibilities:
+
+Handles session lifecycle
+Injects web-search context
+Locks active agent
+Routes to correct agent
+Stores memory
+Returns structured response to UI
+3. Agent Layer
+
+Each agent is modular:
+
+agents/
+    jung_agent.py
+    dream_agent.py
+    shadow_agent.py
+    ...
+
+🧠 AGENT JUNG — CORE DESIGN
+
+Agent Jung is the primary intelligence layer and philosophical anchor of the system.
+
+🎯 Objective
+
+To simulate a Jungian analytical psychologist that:
+
+Interprets user input symbolically
+Encourages introspection
+Guides toward individuation
+Avoids shallow informational responses
+Maintains psychological depth
+⚙️ Functional Pipeline
+1. Input Extraction
+user_input = state.history[-1]["content"]
+2. Web Search (Conditional)
+
+Triggered via heuristic:
+
+should_use_web(query)
+
+Adds:
+
+state.web_context
+3. RAG (Retrieval-Augmented Generation)
+
+From:
+
+/rag/jung/jung.index
+/rag/jung/jung_texts.txt
+
+Using:
+
+FAISS
+sentence-transformers (MiniLM)
+retrieve(query, k=3)
+
+Via:
+
+context_builder.py
+
+Combines:
+
+Jungian knowledge (RAG)
+Real-world context (web search)
+
+5. LLM Prompt Injection
+
+Final prompt includes:
+
+User query
+Retrieved Jungian concepts
+Optional real-world knowledge
+Instructional tone constraints
+
+6. Output Format
+{
+  "type": "question",
+  "agent": "jung",
+  "content": "...",
+  "next_stage": "jung_followup"
+}
+
+🧠 Behavioral Characteristics
+
+Agent Jung is designed to:
+
+Ask reflective questions
+Avoid premature conclusions
+Link ideas to Jungian constructs:
+Persona
+Shadow
+Anima/Animus
+Individuation
+Self
+Integrate cross-disciplinary references:
+Buddhism
+Mythology
+Philosophy
+
+⚠️ Known Issues (Current)
+
+Overuse of greeting phrases
+Occasional verbosity
+Weak grounding when RAG is sparse
+Redundant memory entries (fixed in latest patch)
+Web search not yet deeply integrated into reasoning
+
+🌐 WEB SEARCH TOOL
+
+Located at:
+
+./core/tools/web_search.py
+Uses Tavily API
+Triggered conditionally
+Injected into prompt as auxiliary context
+Logs usage in backend
+
+📚 RAG SYSTEM
+
+Structure
+kb/jung/ → raw text files
+rag/jung/
+    jung.index
+    jung_texts.txt
+Build Process
+python build_index.py
+
+💻 FRONTEND (Vanilla JS)
+
+Interaction Model
+Fire → starts conversation
+Fuel → continues conversation
+Dynamic agent responses
+Per-agent memory panels
+UI Components
+Chat window
+Input box (dynamic placeholder)
+Agent panels (Jung, Dream, etc.)
+Choice buttons (planned)
+End session control
+
+🔁 AGENT MEMORY SYSTEM
+
+Maintains:
+
+window.agentMemory = {
+  jung: [...],
+  dream: [...]
+}
+
+Synced from backend:
+
+agent_log: {
+  agent: "jung",
+  history: [...]
+}
+
+🔮 FUTURE ROADMAP
+
+Immediate
+Remove repetitive greetings (prompt refinement)
+Stabilize RAG grounding
+Clean agent switching UX
+
+Mid-Term
+Multi-agent orchestration
+Synthesizer agent
+Better tool-use reasoning
+
+Advanced
+Fine-tuning (LoRA) on Jungian corpus
+Personality stabilization
+Long-term memory
+
+🧪 DEVELOPMENT NOTES
+
+Built for iterative experimentation
+Modular agent design allows scaling
+RAG per-agent enables specialization
+Web + RAG hybrid = hybrid cognition model
+
+🧭 PHILOSOPHY
+
+CGJI01 is an attempt to model:
+
+"Dialogue as a path to psychological transformation"
+
+Inspired by:
+
+Jungian analytical psychology
+Symbolic reasoning
+Multi-perspective cognition
+
+⚙️ SETUP
+pip install -r requirements.txt
+
+Run:
+
+uvicorn api_server:app --reload

api_server.py

@@ -22,10 +22,8 @@ app = FastAPI()
 # THEN MOUNT STATIC
 app.mount("/static", StaticFiles(directory="web"), name="static")
 
-
 # sessions = {}
 
-
 # ---- CORS FIX ----
 app.add_middleware(
     CORSMiddleware,

config.toml

@@ -0,0 +1,35 @@
+[app]
+name = "CGJI01"
+mode = "development"
+
+[server]
+host = "0.0.0.0"
+port = 8000
+
+[agents]
+default = "jung"
+available = ["jung", "dream", "shadow", "myth", "epistemic", "bpsy", "jred"]
+
+[rag]
+enabled = true
+base_path = "rag/"
+top_k = 3
+
+[web_search]
+enabled = true
+provider = "tavily"
+max_results = 3
+
+[llm]
+provider = "groq"
+model = "llama-3.1-8b-instant"
+temperature = 0.7
+
+[memory]
+per_agent = true
+store_history = true
+
+[ui]
+mode = "interactive"
+conversation_start_label = "Fire"
+conversation_continue_label = "Fuel"

config.yaml

@@ -0,0 +1,48 @@
+
+---
+
+# ⚙️ `config.yaml`
+
+```yaml
+app:
+  name: CGJI01
+  mode: development
+
+server:
+  host: 0.0.0.0
+  port: 8000
+
+agents:
+  default: jung
+  available:
+    - jung
+    - dream
+    - shadow
+    - myth
+    - epistemic
+    - bpsy
+    - jred
+
+rag:
+  enabled: true
+  base_path: rag/
+  top_k: 3
+
+web_search:
+  enabled: true
+  provider: tavily
+  max_results: 3
+
+llm:
+  provider: groq
+  model: llama-3.1-8b-instant
+  temperature: 0.7
+
+memory:
+  per_agent: true
+  store_history: true
+
+ui:
+  mode: interactive
+  conversation_start_label: Fire
+  conversation_continue_label: Fuel