Containerization Support & Dynamic Model Discovery

.env

@@ -0,0 +1,2 @@
+OLLAMA_MODEL=qwen3:30b
+GRAPHRAG_API_KEY=ollama

Dockerfile

@@ -0,0 +1,84 @@
+# Solace-AI: GraphRAG Final Reconciled Dockerfile
+# Optimized for Python 3.12 / Podman / macOS Volume Permissions (2026)
+FROM python:3.12-slim
+
+# 1. Install uv for lightning-fast installs (2026 standard)
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
+
+# Set environment variables for container stability
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    DEBIAN_FRONTEND=noninteractive \
+    PIP_DEFAULT_TIMEOUT=1000 \
+    PIP_RETRIES=10 \
+    POETRY_DYNAMIC_VERSIONING_BYPASS="0.0.0"
+
+WORKDIR /app
+
+# 2. Install system dependencies required for GraphRAG and Git builds
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential curl git ffmpeg && rm -rf /var/lib/apt/lists/*
+
+# 3. Install stable background dependencies using uv
+# Pinned versions preserved exactly from your original requirement
+RUN uv pip install --system --no-cache-dir \
+    gradio==4.44.1 \
+    huggingface_hub==0.23.4 \
+    pydantic==2.7.4 \
+    jinja2==3.1.4 \
+    starlette==0.38.2 \
+    fastapi \
+    uvicorn \
+    python-dotenv \
+    pandas \
+    tiktoken \
+    langchain==0.2.16 \
+    langchain-community==0.2.16 \
+    langchain-core==0.2.38 \
+    aiohttp \
+    pyyaml \
+    requests \
+    duckduckgo-search \
+    ollama \
+    plotly
+
+# 4. Install GraphRAG separately (v0.1.1)
+RUN uv pip install --system --no-cache-dir \
+    git+https://github.com/microsoft/graphrag.git@v0.1.1
+
+# 5. THE PATH FIX: Copy everything from the current directory
+COPY . .
+
+# 6. THE SURGICAL FIXES (All Original Functionality Preserved + Critical Fixes)
+# A. Revert import path for v0.1.1 compatibility
+RUN sed -i 's/graphrag.query.llm.text_utils/graphrag.query.context_builder.entity_extraction/g' api.py
+
+# B. Force FastAPI and Gradio to bind to 0.0.0.0
+RUN sed -i 's/127.0.0.1/0.0.0.0/g' api.py app.py index_app.py
+
+# C. Update Ollama hostname to host.docker.internal for Mac GPU bridge
+RUN sed -i 's/localhost:11434/host.docker.internal:11434/g' app.py index_app.py api.py
+
+# D. Dynamic pathing for the Gradio client patch
+RUN sed -i 's/def _json_schema_to_python_type(schema, defs):/def _json_schema_to_python_type(schema, defs):\n    if isinstance(schema, bool):\n        return "Any"/g' $(python -c "import site; print(site.getsitepackages()[0])")/gradio_client/utils.py
+
+# E. UI PATH DISCOVERY FIX: Hard-code root to /app to resolve "No folder selected"
+RUN sed -i "s|root_dir = os.path.dirname(os.path.abspath(__file__))|root_dir = '/app'|g" app.py
+
+# F. INTERNAL NETWORKING FIX: Force the Indexer to talk to the API container
+# This stops the "Connection refused" error when indexer tries to find localhost
+RUN sed -i "s|localhost:8012|graphrag-api:8012|g" index_app.py || true
+
+# 7. Setup persistent directories and wipe "ghost" build artifacts
+RUN mkdir -p indexing lancedb input cache output
+RUN chmod -R 777 /app
+
+# 8. PERMISSION FIX: Run as root for Podman/macOS Volume compatibility
+# This ensures that the Indexer can write .tmp files and parquet files to your Mac
+USER root
+
+# Expose ports: API (8012), Chat UI (7860), Indexer (7861)
+EXPOSE 8012 7860 7861
+
+# 9. Start the API
+CMD ["python", "api.py", "--host", "0.0.0.0", "--port", "8012"]

app.py

@@ -61,6 +61,35 @@
 import warnings
 warnings.filterwarnings("ignore", category=UserWarning, module="gradio_client.documentation")
 
+# --- PATH & DISCOVERY HELPERS ---
+ROOT_DIR = "/app" if os.path.exists("/app/indexing") else "."
+INDEXING_ROOT = os.path.join(ROOT_DIR, "indexing")
+
+def list_output_folders(root):
+    """Dynamically lists indexing runs in the output folder."""
+    output_dir = os.path.join(root, "output")
+    if not os.path.exists(output_dir):
+        return []
+    return sorted([f for f in os.listdir(output_dir) if os.path.isdir(os.path.join(output_dir, f)) and f != ".DS_Store"], reverse=True)
+
+def fetch_models(base_url, api_key, service_type):
+    """Fetches live models and filters out the 123b/24b strings that cause 404s."""
+    try:
+        if service_type.lower() == "ollama" or "11434" in str(base_url):
+            clean_url = base_url.replace("/v1", "")
+            response = requests.get(f"{clean_url}/api/tags", timeout=5)
+            data = response.json()
+            models = [m['name'] for m in data.get('models', [])]
+        else:
+            response = requests.get(f"{base_url}/v1/models", headers={"Authorization": f"Bearer {api_key}"}, timeout=5)
+            data = response.json()
+            models = [m['id'] for m in data.get('data', [])]
+
+        # PURGE GHOSTS: This stops the 404 error by removing non-existent model strings
+        ghosts = ["mistral-large:123b-instruct-2407-q4_0", "mistral-small:24b"]
+        return [m for m in models if m not in ghosts]
+    except:
+        return []
 
 load_dotenv('indexing/.env')
 
@@ -388,32 +417,26 @@ def parse_query_response(response: str):
         print(f"Error parsing query response: {str(e)}")
         return response 
 
-def send_message(query_type, query, history, system_message, temperature, max_tokens, preset, community_level, response_type, custom_cli_args, selected_folder):
+def send_message(query_type, query, history, system_message, temperature, max_tokens, preset, community_level, response_type, custom_cli_args, selected_folder, selected_model):
     try:
         if query_type in ["global", "local"]:
             cli_args = construct_cli_args(query_type, preset, community_level, response_type, custom_cli_args, query, selected_folder)
-            logging.info(f"Executing {query_type} search with command: {' '.join(cli_args)}")
             result = run_graphrag_query(cli_args)
             parsed_result = parse_query_response(result)
-            logging.info(f"Parsed query result: {parsed_result}")
         else:  # Direct chat
-            llm_model = os.getenv("LLM_MODEL")
             api_base = os.getenv("LLM_API_BASE")
-            logging.info(f"Executing direct chat with model: {llm_model}")
-
+            # THE FIX: Use 'selected_model' from the UI instead of os.getenv
+            logging.info(f"Executing direct chat with live selection: {selected_model}")
             try:
-                result = chat_with_llm(query, history, system_message, temperature, max_tokens, llm_model, api_base)
-                parsed_result = result  # No parsing needed for direct chat
-                logging.info(f"Direct chat result: {parsed_result[:100]}...")  # Log first 100 chars of result
+                result = chat_with_llm(query, history, system_message, temperature, max_tokens, selected_model, api_base)
+                parsed_result = result
             except Exception as chat_error:
-                logging.error(f"Error in chat_with_llm: {str(chat_error)}")
                 raise RuntimeError(f"Direct chat failed: {str(chat_error)}")
 
         history.append((query, parsed_result))
     except Exception as e:
         error_message = f"An error occurred: {str(e)}"
         logging.error(error_message)
-        logging.exception("Exception details:")
         history.append((query, error_message))
 
     return history, gr.update(value=""), update_logs()
@@ -1554,9 +1577,7 @@ def create_gradio_interface():
                         )
                         selected_folder = gr.Dropdown(
                             label="Select Index Folder to Chat With",
-                            choices=list_output_folders("./indexing"),
-                            value=None,
-                            interactive=True
+                            choices=list_output_folders(ROOT_DIR)
                         )
                         refresh_folder_btn = gr.Button("Refresh Folders", variant="secondary")
                         clear_chat_btn = gr.Button("Clear Chat", variant="secondary")
@@ -1703,7 +1724,8 @@ def update_custom_options(preset):
                 community_level,
                 response_type,
                 custom_cli_args,
-                selected_folder
+                selected_folder,
+                llm_model_dropdown
             ],
             outputs=[chatbot, query_input, log_output]
         )
@@ -1721,7 +1743,8 @@ def update_custom_options(preset):
                 community_level,
                 response_type,
                 custom_cli_args,
-                selected_folder
+                selected_folder,
+                llm_model_dropdown
             ],
             outputs=[chatbot, query_input, log_output]
         )

docker-compose.yml

@@ -0,0 +1,66 @@
+# cbt-graph-rag-containerised: Dynamic Model Discovery (2026)
+version: '3.8'
+
+services:
+  graphrag-api:
+    build: .
+    container_name: graphrag-api
+    environment:
+      - PYTHONPATH=/app
+      - GRAPHRAG_API_KEY=ollama
+      - OLLAMA_BASE_URL=http://host.docker.internal:11434/v1
+    ports:
+      - "8012:8012"
+    volumes:
+      - ./input:/app/input
+      - ./indexing:/app/indexing
+      - ./lancedb:/app/lancedb
+      - ./indexing/output:/app/output
+      - ./indexing/output:/app/indexing/output
+    command: python api.py --host 0.0.0.0 --port 8012
+
+  graphrag-ui:
+    build: .
+    container_name: graphrag-ui
+    environment:
+      - PYTHONPATH=/app
+      - GRAPHRAG_API_KEY=ollama
+      # THE FIX: Point UI to the API service name
+      - API_URL=http://graphrag-api:8012 
+      - GRADIO_SERVER_NAME=0.0.0.0
+      - LLM_API_BASE=http://host.docker.internal:11434/v1
+      - EMBEDDINGS_API_BASE=http://host.docker.internal:11434/v1
+    ports:
+      - "7860:7860"
+    volumes:
+      - .:/app # Mount the whole root so .env updates are persisted to your Mac
+      - ./indexing/output:/app/output
+      - ./indexing/output:/app/indexing/output
+    depends_on:
+      - graphrag-api
+    command: python app.py
+
+  graphrag-indexer:
+    build: .
+    container_name: graphrag-indexer
+    environment:
+      - PYTHONPATH=/app
+      - GRAPHRAG_API_KEY=ollama
+      - API_URL=http://graphrag-api:8012 
+      - GRADIO_SERVER_NAME=0.0.0.0
+      - LLM_API_BASE=http://host.docker.internal:11434/v1
+      - EMBEDDINGS_API_BASE=http://host.docker.internal:11434/v1
+    ports:
+      - "7861:7861"
+    volumes:
+      - .:/app
+      - ./indexing/output:/app/output
+      - ./indexing/output:/app/indexing/output
+    depends_on:
+      - graphrag-api
+    command: python index_app.py
+
+networks:
+  default:
+    name: graphrag-network
+    driver: bridge

indexing/.env

@@ -1,11 +1,11 @@
 LLM_PROVIDER=openai
-LLM_API_BASE=http://localhost:11434/v1
-LLM_MODEL='mistral-large:123b-instruct-2407-q4_0'
+LLM_API_BASE=http://host.docker.internal:11434/v1
+LLM_MODEL=mistral-small:24b
 LLM_API_KEY=12345
 
 EMBEDDINGS_PROVIDER=openai
-EMBEDDINGS_API_BASE=http://localhost:11434
-EMBEDDINGS_MODEL='snowflake-arctic-embed:335m'
+EMBEDDINGS_API_BASE=http://host.docker.internal:11434/v1
+EMBEDDINGS_MODEL=nomic-embed-text:latest
 EMBEDDINGS_API_KEY=12345
 
 
@@ -17,3 +17,7 @@ EMBEDDINGS_SERVICE_TYPE=openai_embedding
 
 API_URL=http://localhost:8012
 API_PORT=8012
+CONTEXT_WINDOW=4096
+SYSTEM_MESSAGE=You are a helpful AI assistant.
+TEMPERATURE=0.5
+MAX_TOKENS=1024

indexing/input/MSA_Novalytica_Distributor.txt

@@ -0,0 +1,15 @@
+MASTER SOFTWARE RESELLER AGREEMENT
+
+EFFECTIVE DATE: January 1, 2025
+
+PARTIES: Novalytica, Inc. ("Vendor") and Global Tech Distribution Corp ("Distributor")
+
+1. APPOINTMENT > Vendor hereby appoints Distributor as a non-exclusive distributor of the Vendor's enterprise software products within the agreed territory.
+
+2. BUNDLING AND RESALE > Distributor may grant downstream System Integrators (SIs) the right to bundle Vendor software with authorized hardware and services.
+
+3. COMPLIANCE AND RESTRICTIONS > 3.1 General Compliance: Distributor and its authorized SIs shall comply with all applicable local and international trade laws.
+
+3.2 Government Sales Restriction: Resale of enterprise software licenses to municipal or national government entities is strictly prohibited without 30 days prior written consent from Novalytica, Inc. Failure to obtain consent will result in immediate termination of the downstream license and potential liability for the Distributor.
+
+4. TERM AND TERMINATION > This agreement shall remain in effect for three (3) years from the Effective Date unless terminated earlier due to a material breach.

indexing/input/PO_SingaporeLTA_Acme.txt

@@ -0,0 +1,27 @@
+OFFICIAL PURCHASE ORDER
+
+PO NUMBER: LTA-2026-88492
+
+DATE OF ISSUE: April 2, 2026
+
+BUYER: > Singapore Land Transport Authority (LTA)
+
+1 Hampshire Road
+
+Singapore 219428
+
+VENDOR: > Acme Systems LLC
+
+Enterprise Solutions Division
+
+LINE ITEMS: > 1. Custom AI Traffic Analytics Dashboard - Qty: 1 - $125,000
+
+2. High-Performance Compute Cluster Setup - Qty: 1 - $85,000
+
+3. Data Analytics Enterprise Software Bundle (Includes Quantus AI Studio & Novalytica Enterprise License) - Qty: 1 - $150,000
+
+4. 24/7 Managed Support Services (1 Year) - Qty: 1 - $40,000
+
+TOTAL: $400,000 USD
+
+TERMS: Net 30. Delivery expected by May 15, 2026.

indexing/input/Partner_Agreement_AcmeSystems.txt

@@ -0,0 +1,17 @@
+SYSTEM INTEGRATOR PARTNER AGREEMENT
+
+EFFECTIVE DATE: June 15, 2025
+
+PARTIES: Global Tech Distribution Corp ("Distributor") and Acme Systems LLC ("System Integrator")
+
+1. PARTNER STATUS > Distributor grants Acme Systems LLC the status of "Authorized Platinum Integrator."
+
+2. AUTHORIZED PORTFOLIO > Acme Systems LLC is hereby authorized to procure, bundle, and resell the following software platforms provided by the Distributor:
+
+Quantus AI Studio
+
+Novalytica Enterprise
+
+NexusSearch Cloud
+
+3. LIABILITY AND INDEMNIFICATION > Acme Systems LLC assumes full responsibility for front-line deployment and integration services. Distributor shall not be held liable for SLA breaches caused by Acme Systems LLC's custom integrations. Acme Systems LLC must adhere to all original Vendor MSAs passed through the Distributor.

indexing/output/20260415-034514/artifacts/stats.json

@@ -0,0 +1,30 @@
+{
+    "total_runtime": 7.076313495635986,
+    "num_documents": 4,
+    "input_load_time": 0,
+    "workflows": {
+        "create_base_text_units": {
+            "overall": 6.424093008041382,
+            "0_orderby": 0.00046443939208984375,
+            "1_zip": 0.0004918575286865234,
+            "2_aggregate_override": 0.0011968612670898438,
+            "3_chunk": 6.416795015335083,
+            "4_select": 0.0006856918334960938,
+            "5_unroll": 0.0008931159973144531,
+            "6_rename": 0.0003285408020019531,
+            "7_genid": 0.0008454322814941406,
+            "8_unzip": 0.0004532337188720703,
+            "9_copy": 0.00020647048950195312,
+            "10_filter": 0.0015630722045898438
+        },
+        "create_base_extracted_entities": {
+            "overall": 0.38549017906188965,
+            "0_entity_extract": 0.3832550048828125,
+            "1_merge_graphs": 0.002129077911376953
+        },
+        "create_summarized_entities": {
+            "overall": 0.0012483596801757812,
+            "0_summarize_descriptions": 0.0011658668518066406
+        }
+    }
+}