Switch LLM provider from Gemini to OpenAI GPT-5.4-mini

The live app ran the Gemini provider with no fallback, so any key/quota
issue made chat silently fail with no response. Switch the orchestrator
and personas to OpenAI gpt-5.4-mini:

- bootstrap.py: add an explicit "openai" provider branch so the OpenAI
client is used as the primary LLM (not just a vLLM fallback).
- muscle_growth_config.yaml: llm.provider=openai, openai.model=gpt-5.4-mini.
- .env.example / README: document OPENAI_API_KEY as the primary secret;
keep GEMINI_API_KEY as an optional fallback.

Key is supplied via the OPENAI_API_KEY env var / HF Space secret and is
never committed. Verified locally end-to-end (signup + chat-stream) with
all personas returning responses.

Co-authored-by: Cursor <cursoragent@cursor.com>

README.md

@@ -8,36 +8,102 @@ pinned: false
 app_port: 7860
 ---
 
-# MuscleGrowthAI
+# MuscleGrowthAI Panel
 
-An AI-powered bodybuilding advisor panel that helps gym-goers build personalized muscle-building plans. Get guidance on hypertrophy programming, nutrition, recovery, form, and progress tracking from five specialized fitness AI advisors. Built on Neon AI's Collaborative Conversational AI (CCAI) framework.
+An AI personalized bodybuilding assistant built on Neon AI's Collaborative
+Conversational AI (CCAI) framework. Ask about hypertrophy programming,
+nutrition, recovery, form, and progress tracking and get diverse perspectives
+from a panel of five fitness AI advisors.
+
+This repository is a **complete, deployable application** — the CCAI
+multi-advisor stack (FastAPI backend + React frontend) wired to the
+MuscleGrowthAI configuration in [`muscle_growth_config.yaml`](muscle_growth_config.yaml)
+and the personas in [`personas/fitness_advisors/`](personas/fitness_advisors).
+
+## Advisors
+
+1. **Hypertrophy Coach** — splits, sets/reps, muscle-group programming
+2. **Nutrition Strategist** — protein, macros, meal timing
+3. **Recovery Specialist** — rest, stretching, soreness management
+4. **Form & Safety Coach** — technique, breathing, injury prevention
+5. **Program Planner** — scheduling, tracking, progression
 
 ## Hugging Face Spaces deployment
 
-This Space ships as a single Docker image built from the repository root [`Dockerfile`](Dockerfile). The container:
+This Space ships as a single Docker image built from the repository-root
+[`Dockerfile`](Dockerfile). The container:
 
-1. Builds the React frontend at image-build time with `REACT_APP_API_URL=""` so every `fetch` uses relative URLs.
-2. Serves the bundled SPA from FastAPI at `/`, with the API on `/api/...`, `/auth/...`, etc., on the same `:7860` origin.
-3. Persists user data in **SQLite via `aiosqlite`** at `${DATA_DIR}/` (default `/home/user/data` in the Docker image). Mount a Hugging Face Storage Bucket there to survive Space rebuilds.
+1. Builds the React frontend (CRA) at image-build time with `REACT_APP_API_URL=""`
+   so every `fetch` issues a relative URL.
+2. Serves the bundled SPA from FastAPI at `/`, with the API on `/api/...`,
+   `/auth/...`, etc. — all on the same `:7860` origin.
+3. Persists user data (auth, profiles, chat sessions) in **SQLite via
+   `aiosqlite`** at `${DATA_DIR}/muscle_growth_panel.db`. Mount a Hugging Face
+   Storage Bucket at `/data` to make the database survive Space rebuilds.
+   There is **no MongoDB** and no third-party data plane.
 
 ### Required Space secrets
 
 | Secret | Purpose |
 |--------|---------|
-| `JWT_SECRET_KEY` | Signs auth tokens. Set to a long random string. |
-| `GEMINI_API_KEY` | Powers the default Gemini provider (`gemini-2.5-flash`). |
+| `JWT_SECRET_KEY` | Signs auth tokens. Set this to a long random string. |
+| `OPENAI_API_KEY` | Powers the default OpenAI provider (`gpt-5.4-mini`). Get a key from Clary, or use your own OpenAI key. |
+| `GEMINI_API_KEY` | Optional — only if you switch `llm.provider` back to `gemini` (`gemini-2.5-flash`). |
 
-### Optional secrets
+Set these under **Settings → Variables and secrets** on the Space.
 
-| Secret | Purpose |
-|--------|---------|
-| `OPENAI_API_KEY` | Only if switching to the OpenAI provider. |
-| `VLLM_API_KEY` | Only if pointing at a Neon vLLM endpoint. |
+## Local deployment
 
-## Advisors
+**Do you need Docker?** No — Docker is optional. There are two supported paths,
+and neither requires MongoDB (persistence is SQLite):
 
-1. **Hypertrophy Coach** — splits, sets/reps, programming
-2. **Nutrition Strategist** — protein, macros, fueling
-3. **Recovery Specialist** — rest, mobility, post-workout care
-4. **Form & Safety Coach** — technique, breathing, injury prevention
-5. **Program Planner** — scheduling, tracking, progression
+### Option A — Docker (simplest, mirrors the Space exactly)
+
+Requires **Docker Desktop** only.
+
+```bash
+# From the repo root, create a .env with at least:
+#   JWT_SECRET_KEY=some-long-random-string
+#   OPENAI_API_KEY=your-openai-key
+docker compose up --build
+```
+
+Open <http://localhost:7860>. Override the host port with `MUSCLE_HOST_PORT` if
+7860 is taken.
+
+### Option B — Native (no Docker)
+
+Requires **Python 3.12** and **Node.js 20+** (no Docker, no MongoDB).
+
+**Backend** (terminal 1):
+
+```bash
+cd multi_llm_chatbot_backend
+python -m venv venv
+# Windows: venv\Scripts\activate   •   macOS/Linux: source venv/bin/activate
+pip install -r requirements.txt
+cp .env.example .env         # then edit JWT_SECRET_KEY + OPENAI_API_KEY
+uvicorn app.main:app --reload --port 8000
+```
+
+**Frontend** (terminal 2):
+
+```bash
+cd phd-advisor-frontend
+npm install
+# point the SPA at the backend from step above:
+#   Windows PowerShell:  $env:REACT_APP_API_URL="http://localhost:8000"; npm start
+#   macOS/Linux:         REACT_APP_API_URL=http://localhost:8000 npm start
+npm start
+```
+
+Open <http://localhost:3000> — you should see **AI Personalized Bodybuilding
+Plan** with the five fitness advisors.
+
+## Configuration
+
+All branding, login fields, chat examples, orchestrator keywords, and LLM/RAG
+settings live in [`muscle_growth_config.yaml`](muscle_growth_config.yaml). The
+backend resolves `personas.personas_dir` relative to that file, so the advisor
+YAMLs in `personas/fitness_advisors/` load automatically. Point the app at a
+different config with the `CONFIG_PATH` environment variable.

multi_llm_chatbot_backend/.env.example

@@ -1,16 +1,24 @@
-# Path to app config (use cybersecurity_config.yaml for this branch)
-CONFIG_PATH=../cybersecurity_config.yaml
+# Copy this file to `.env` (same directory) for a native, non-Docker local run.
+# The backend auto-loads it on startup.
 
-MONGODB_CONNECTION_STRING=mongodb://localhost:27017
-JWT_SECRET_KEY=change-me
+# Path to the app config, relative to this backend directory.
+CONFIG_PATH=../muscle_growth_config.yaml
 
-# From C:\Users\dream\.secrets\shared.env or set here:
-VLLM_API_KEY=
+# Where the SQLite database + uploaded docs live. Any writable folder works;
+# on a native run this replaces the container's /data mount.
+DATA_DIR=./local_data
+
+# Signs auth tokens — set to a long random string.
+JWT_SECRET_KEY=change-me-to-a-long-random-string
+
+# Default provider is OpenAI GPT-5.4-mini (see muscle_growth_config.yaml -> llm.provider).
+# Set your OpenAI API key here (get one from Clary, or use your own).
 OPENAI_API_KEY=
-GEMINI_API_KEY=
 
-# Include every frontend origin you use locally (React default is 3000)
-CORS_ORIGINS=http://localhost:3010,http://127.0.0.1:3010,http://localhost:3000,http://127.0.0.1:3000
+# Optional alternatives:
+# GEMINI_API_KEY=   # only if you switch llm.provider back to "gemini"
+VLLM_API_KEY=
 
-# Optional override; default is %USERPROFILE%\.secrets\shared.env
-# SHARED_ENV=C:\Users\dream\.secrets\shared.env
+# Frontend origins allowed to call this backend during native dev
+# (CRA dev server defaults to port 3000).
+CORS_ORIGINS=http://localhost:3000,http://127.0.0.1:3000

multi_llm_chatbot_backend/app/core/bootstrap.py

@@ -18,7 +18,7 @@ from app.models.default_personas import get_default_personas
 settings = get_settings()
 
 current_provider = settings.llm.provider or "vllm"
-available_providers = ["ollama", "gemini", "vllm"]
+available_providers = ["ollama", "gemini", "vllm", "openai"]
 
 
 def _load_shared_env_var(name: str) -> str:
@@ -114,6 +114,11 @@ def _wrap_resilient(primary: ImprovedVllmClient, fallback: OpenAIFallbackClient,
 def create_llm_client(provider=None):
     if provider is None:
         provider = current_provider
+    if provider == "openai":
+        # Standalone OpenAI (GPT) client. OpenAIFallbackClient implements the same
+        # LLMClient interface as the Gemini client, so it works as the primary
+        # orchestrator/persona LLM, not just a vLLM fallback.
+        return _build_openai(settings.llm.openai.orchestrator_reasoning_effort)
     if provider == "gemini":
         return ImprovedGeminiClient(model_name=settings.llm.gemini.model)
     if provider == "vllm":

muscle_growth_config.yaml

@@ -169,7 +169,7 @@ mongodb:
   database_name: "muscle_growth_advisor"
 
 llm:
-  provider: "gemini"
+  provider: "openai"
   gemini:
     model: "gemini-2.5-flash"
   ollama:
@@ -182,8 +182,9 @@ llm:
     neon_persona_orchestrator: "vanilla"
     neon_persona_advisors: "vanilla"
   openai:
+    # Key comes from the OPENAI_API_KEY env var / HF Space secret (never committed).
     api_key: ""
-    model: "gpt-5.4"
+    model: "gpt-5.4-mini"
     orchestrator_reasoning_effort: "low"
     persona_reasoning_effort: "none"
   resilient: