sunholo 0.60.1__tar.gz → 0.60.3__tar.gz

{sunholo-0.60.1 → sunholo-0.60.3}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: sunholo
-Version: 0.60.1
+Version: 0.60.3
 Summary: Large Language Model DevOps - a package to help deploy LLMs to the Cloud.
 Home-page: https://github.com/sunholo-data/sunholo-py
-Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.60.1.tar.gz
+Download-URL: https://github.com/sunholo-data/sunholo-py/archive/refs/tags/v0.60.3.tar.gz
 Author: Holosun ApS
 Author-email: multivac@sunholo.com
 License: Apache License, Version 2.0
@@ -18,13 +18,13 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Description-Content-Type: text/markdown
 License-File: LICENSE.txt
-Requires-Dist: jsonschema
+Requires-Dist: google-auth
 Requires-Dist: langchain
 Requires-Dist: langchain_experimental
 Requires-Dist: langchain-community
-Requires-Dist: google-auth
 Provides-Extra: all
 Requires-Dist: asyncpg; extra == "all"
+Requires-Dist: fastapi; extra == "all"
 Requires-Dist: flask; extra == "all"
 Requires-Dist: google-auth; extra == "all"
 Requires-Dist: google-auth-httplib2; extra == "all"
@@ -42,6 +42,7 @@ Requires-Dist: google-generativeai; extra == "all"
 Requires-Dist: gunicorn; extra == "all"
 Requires-Dist: httpcore; extra == "all"
 Requires-Dist: httpx; extra == "all"
+Requires-Dist: jsonschema; extra == "all"
 Requires-Dist: lancedb; extra == "all"
 Requires-Dist: langchain; extra == "all"
 Requires-Dist: langchain_experimental; extra == "all"
@@ -55,10 +56,13 @@ Requires-Dist: pg8000; extra == "all"
 Requires-Dist: pgvector; extra == "all"
 Requires-Dist: psycopg2-binary; extra == "all"
 Requires-Dist: pypdf; extra == "all"
-Requires-Dist: fastapi; extra == "all"
+Requires-Dist: python-socketio; extra == "all"
+Requires-Dist: rich; extra == "all"
 Requires-Dist: supabase; extra == "all"
 Requires-Dist: tiktoken; extra == "all"
-Requires-Dist: python-socketio; extra == "all"
+Provides-Extra: cli
+Requires-Dist: jsonschema; extra == "cli"
+Requires-Dist: rich; extra == "cli"
 Provides-Extra: database
 Requires-Dist: asyncpg; extra == "database"
 Requires-Dist: supabase; extra == "database"

{sunholo-0.60.1 → sunholo-0.60.3}/setup.py

@@ -1,7 +1,7 @@
 from setuptools import setup, find_packages
 
 # Define your base version
-version = '0.60.1'
+version = '0.60.3'
 
 setup(
     name='sunholo',
@@ -28,17 +28,17 @@ setup(
     },
     install_requires=[
         # Base dependencies
-        "jsonschema",
+        "google-auth", # to check if on gcp
         "langchain",
         "langchain_experimental",
         "langchain-community",
-        "google-auth" # to check if on gcp
         # Add the minimal dependencies that your package requires here
     ],
     extras_require={
         # Define optional dependencies with feature names
         'all': [
             "asyncpg",
+            "fastapi",
             "flask",
             "google-auth",
             "google-auth-httplib2",
@@ -56,6 +56,7 @@ setup(
             "gunicorn",
             "httpcore",
             "httpx",
+            "jsonschema",
             "lancedb",
             "langchain",
             "langchain_experimental",
@@ -69,10 +70,15 @@ setup(
             "pgvector",
             "psycopg2-binary",
             "pypdf",
-            "fastapi",
+            "python-socketio",
+            "rich",
             "supabase",
             "tiktoken",
-            "python-socketio"
+
+        ],
+        'cli': [
+            "jsonschema",
+            "rich"
         ],
         'database': [
             "asyncpg",

{sunholo-0.60.1 → sunholo-0.60.3}/sunholo/__init__.py

@@ -19,6 +19,7 @@ from . import utils
 from . import vertex
 import logging
 
+
 __all__ = ['agents', 
            'archive', 
            'auth',
@@ -37,5 +38,7 @@ __all__ = ['agents',
            'qna', 
            'streaming', 
            'utils', 
+           'vertex',
            'logging']
 
+

{sunholo-0.60.1 → sunholo-0.60.3}/sunholo/agents/dispatch_to_qa.py

@@ -52,7 +52,7 @@ def prep_request_payload(user_input, chat_history, vector_name, stream, **kwargs
     agent = load_config_key("agent", vector_name=vector_name, kind="vacConfig")
     agent_type = load_config_key("agent_type", vector_name=vector_name, kind="vacConfig")
 
-    override_endpoint = kwargs.get("endpoint_url")
+    override_endpoint = kwargs.get("override_endpoint")
     if override_endpoint:
         log.info(f"Overriding endpoint with {override_endpoint}")
 

{sunholo-0.60.1 → sunholo-0.60.3}/sunholo/agents/flask/qna_routes.py

@@ -37,6 +37,10 @@ except ImportError as err:
 
 def register_qna_routes(app, stream_interpreter, vac_interpreter):
 
+    @app.route("/")
+    def home():
+        return jsonify("OK")
+
     @app.route('/vac/streaming/<vector_name>', methods=['POST'])
     @observe()
     def stream_qa(vector_name):

{sunholo-0.60.1 → sunholo-0.60.3}/sunholo/cli/chat_vac.py

@@ -1,38 +1,46 @@
-from ..agents import send_to_qa_async
-from ..streaming import generate_proxy_stream_async, generate_proxy_stream
-from ..utils.user_ids import generate_uuid_from_gcloud_user
+from ..agents import send_to_qa
+from ..streaming import generate_proxy_stream
+from ..utils.user_ids import generate_user_id
+from ..utils.config import load_config_key
 
-from .run_proxy import load_proxies, start_proxy
+from .run_proxy import clean_proxy_list, start_proxy
 
 import uuid
 
-def get_service_url(service_name):
-    proxies = load_proxies()
+from rich import print
+from .sun_rich import console
+
+from rich.prompt import Prompt
+from rich.panel import Panel
+from rich.text import Text
+
+
+def get_service_url(service_name, project, region):
+    proxies = clean_proxy_list()
     if service_name in proxies:
         port = proxies[service_name]['port']
         return f"http://127.0.0.1:{port}"
     else:
         print(f"No proxy found running for service: {service_name} - attempting to connect")
-        return start_proxy(service_name)
+        return start_proxy(service_name, region, project)
 
-async def stream_chat_session(service_name, chat_history):
+def stream_chat_session(service_name, project, region):
 
-    service_url = get_service_url(service_name)
-    user_id = generate_uuid_from_gcloud_user()
+    service_url = get_service_url(service_name, project, region)
+    user_id = generate_user_id()
+    chat_history = []
     while True:
         session_id = str(uuid.uuid4())
-        user_input = input("You: ")
+        user_input = Prompt.ask("[bold cyan]You[/bold cyan]")
         if user_input.lower() in ["exit", "quit"]:
-            print("Exiting chat session.")
+            console.print("[bold red]Exiting chat session.[/bold red]")
             break
 
-        chat_history.append({"role": "user", "content": user_input})
-
-        
+        chat_history.append({"role": "Human", "content": user_input})
 
-        async def stream_response():
-            generate = await generate_proxy_stream_async(
-                send_to_qa_async,
+        def stream_response():
+            generate = generate_proxy_stream(
+                send_to_qa,
                 user_input,
                 vector_name=service_name,
                 chat_history=chat_history,
@@ -56,28 +64,40 @@ async def stream_chat_session(service_name, chat_history):
                 message_source="cli",
                 override_endpoint=service_url
             )
-            async for part in generate():
+            for part in generate():
                 yield part
 
-        print("Assistant: ", end='', flush=True)
-        async for token in stream_response():
-            if isinstance(token, bytes):
-                token = token.decode('utf-8')
-            print(token, end='', flush=True)
+        response_started = False
+        vac_response = ""
 
-        chat_history.append({"role": "assistant", "content": token})
-        print()  # For new line after streaming ends
+        # point or star?
+        with console.status("[bold orange]Thinking...[/bold orange]", spinner="star") as status:
+            for token in stream_response():
+                if not response_started:
+                    status.stop()
+                    console.print(f"[bold yellow]{service_name}:[/bold yellow] ", end='')
+                    response_started = True
 
-async def headless_mode(service_name, user_input, chat_history=None):
+                if isinstance(token, bytes):
+                    token = token.decode('utf-8')
+                console.print(token, end='')
+                vac_response += token
+
+        chat_history.append({"role": "AI", "content": vac_response})
+        response_started = False
+        console.print()
+        console.rule()
+
+def headless_mode(service_name, user_input, project, region, chat_history=None):
     chat_history = chat_history or []
-    chat_history.append({"role": "user", "content": user_input})
-    service_url = get_service_url(service_name)
-    user_id = generate_uuid_from_gcloud_user()
+    chat_history.append({"role": "Human", "content": user_input})
+    service_url = get_service_url(project, region)
+    user_id = generate_user_id()
     session_id = str(uuid.uuid4())
 
-    async def stream_response():
-        generate = await generate_proxy_stream_async(
-                send_to_qa_async,
+    def stream_response():
+        generate = generate_proxy_stream(
+                send_to_qa,
                 user_input,
                 vector_name=service_name,
                 chat_history=chat_history,
@@ -101,30 +121,41 @@ async def headless_mode(service_name, user_input, chat_history=None):
                 message_source="cli",
                 override_endpoint=service_url
         )
-        async for part in generate():
+        for part in generate():
             yield part
 
-    print("Assistant: ", end='', flush=True)
-    async for token in stream_response():
+    print(f"VAC {service_name}: ", end='', flush=True)
+    for token in stream_response():
         if isinstance(token, bytes):
             token = token.decode('utf-8')
         print(token, end='', flush=True)
 
-    chat_history.append({"role": "assistant", "content": token})
+    chat_history.append({"role": "AI", "content": token})
     print()  # For new line after streaming ends
 
+    return chat_history
+
 
 def vac_command(args):
     try:
-        service_url = get_service_url(args.service_name)
+        service_url = get_service_url(args.service_name, args.project, args.region)
     except ValueError as e:
-        print(e)
+        console.print(f"[bold red]ERROR: Could not start {args.service_name} proxy URL: {str(e)}[/bold red]")
         return
-
+    
+    display_name = load_config_key("display_name", vector_name=args.service_name,  kind="vacConfig")
+    description = load_config_key("description", vector_name=args.service_name, kind="vacConfig")
+
+    print(
+        Panel(description or "Starting VAC chat session", 
+              title=display_name or args.service_name,
+              subtitle=service_url)
+              )
+    
     if args.headless:
-        headless_mode(service_url, args.user_input, args.chat_history)
+        headless_mode(args.service_name, args.user_input, args.project, args.region, args.chat_history)
     else:
-        stream_chat_session(service_url)
+        stream_chat_session(args.service_name, args.project, args.region)
 
 def setup_vac_subparser(subparsers):
     """
@@ -133,9 +164,9 @@ def setup_vac_subparser(subparsers):
     Args:
         subparsers: The subparsers object from argparse.ArgumentParser().
     """
-    vac_parser = subparsers.add_parser('vac', help='Interact with the VAC service.')
+    vac_parser = subparsers.add_parser('vac', help='Interact with deployed VAC services.')
     vac_parser.add_argument('service_name', help='Name of the VAC service.')
-    vac_parser.add_argument('user_input', help='User input for the VAC service.', nargs='?', default=None)
+    vac_parser.add_argument('user_input', help='User input for the VAC service when in headless mode.', nargs='?', default=None)
     vac_parser.add_argument('--headless', action='store_true', help='Run in headless mode.')
     vac_parser.add_argument('--chat_history', help='Chat history for headless mode (as JSON string).', default=None)
     vac_parser.set_defaults(func=vac_command)

{sunholo-0.60.1 → sunholo-0.60.3}/sunholo/cli/cli.py

@@ -1,4 +1,5 @@
 import argparse
+import logging
 
 from .configs import setup_list_configs_subparser
 from .deploy import setup_deploy_subparser
@@ -6,19 +7,44 @@ from .cli_init import setup_init_subparser
 from .merge_texts import setup_merge_text_subparser
 from .run_proxy import setup_proxy_subparser
 from .chat_vac import setup_vac_subparser
+from ..utils.config import load_config_key
 
+from ..logging import log
+
+from rich import print
+from rich.console import Console
+from rich.prompt import Prompt
+from rich.spinner import Spinner
+from rich.panel import Panel
+from rich.text import Text
+console = Console()
+
+def load_default_gcp_config():
+    gcp_config = load_config_key('gcp_config', 'global', kind="vacConfig")
+    if gcp_config:
+        return gcp_config.get('project_id', ''), gcp_config.get('location', 'europe-west1')
+    else:
+        return '', 'europe-west1'
 
 def main(args=None):
+
+
     """
     Entry point for the sunholo console script. This function parses command line arguments
     and invokes the appropriate functionality based on the user input.
 
-    Example commands:
+    Get started:
     ```bash
-    sunholo deploy --config_path . --gcs_bucket your-gcs-bucket --lancedb_bucket your-lancedb-bucket
+    sunholo --help
     ```
     """
+    default_project, default_region = load_default_gcp_config()
+
     parser = argparse.ArgumentParser(description="sunholo CLI tool for deploying GenAI VACs")
+    parser.add_argument('--debug', action='store_true', help='Enable debug output')
+    parser.add_argument('--project', default=default_project, help='GCP project to list Cloud Run services from.')
+    parser.add_argument('--region', default=default_region, help='Region to list Cloud Run services from.')
+    
     subparsers = parser.add_subparsers(title='commands', 
                                        description='Valid commands', 
                                        help='Commands', 
@@ -40,6 +66,13 @@ def main(args=None):
 
     args = parser.parse_args(args)
 
+    if args.debug:
+        log.setLevel(logging.INFO)
+        logging.getLogger().setLevel(logging.INFO)
+    else:
+        log.setLevel(logging.WARNING)
+        logging.getLogger().setLevel(logging.WARNING)
+
     if hasattr(args, 'func'):
         args.func(args)
     else:

{sunholo-0.60.1 → sunholo-0.60.3}/sunholo/cli/run_proxy.py

@@ -3,6 +3,10 @@ import os
 import signal
 import json
 
+from .sun_rich import console
+from rich.table import Table
+from rich import print
+
 PROXY_TRACKER_FILE = '.vac_proxy_tracker.json'
 DEFAULT_PORT = 8080
 
@@ -46,13 +50,13 @@ def check_gcloud():
     """
     try:
         # Check if gcloud is installed
-        result = subprocess.run(["gcloud", "--version"], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        result = subprocess.run(["gcloud", "--version"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=30)
         if result.returncode != 0:
             print("ERROR: gcloud is not installed or not found in PATH.")
             return False
 
         # Check if gcloud is authenticated
-        result = subprocess.run(["gcloud", "auth", "list"], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        result = subprocess.run(["gcloud", "auth", "list"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=30)
         if result.returncode != 0 or "ACTIVE" not in result.stdout.decode():
             print("ERROR: gcloud is not authenticated. Please run 'gcloud auth login'.")
             return False
@@ -63,16 +67,35 @@ def check_gcloud():
         print(f"ERROR: An unexpected error occurred: {e}")
         return False
 
+def is_process_running(pid):
+    try:
+        os.kill(pid, 0)
+        return True
+    except OSError:
+        print("WARNING: VAC Proxy lost connection")
+        return False
+
 def load_proxies():
     if os.path.exists(PROXY_TRACKER_FILE):
         with open(PROXY_TRACKER_FILE, 'r') as file:
             return json.load(file)
     return {}
 
+def clean_proxy_list():
+    proxies = load_proxies()
+    updated_proxies = {k: v for k, v in proxies.items() if is_process_running(v["pid"])}
+    if len(proxies) != len(updated_proxies):
+        save_proxies(updated_proxies)
+        
+    return updated_proxies
+
 def save_proxies(proxies):
     with open(PROXY_TRACKER_FILE, 'w') as file:
         json.dump(proxies, file, indent=4)
 
+
+
+
 def start_proxy(service_name, region, project, port=None):
     """
     Starts the gcloud proxy to the Cloud Run service and stores the PID.
@@ -83,7 +106,7 @@ def start_proxy(service_name, region, project, port=None):
         project (str): GCP project of the Cloud Run service.
         port (int, optional): Port to run the proxy on. If not provided, auto-assigns the next available port.
     """
-    proxies = load_proxies()
+    proxies = clean_proxy_list()
 
     if service_name in proxies:
         print(f"Proxy for service {service_name} is already running on port {proxies[service_name]['port']}.")
@@ -109,7 +132,7 @@ def start_proxy(service_name, region, project, port=None):
     
     print(f"Proxy for {service_name} setup complete on port {port}")
     list_proxies()
-    
+
     return f"http://127.0.0.1:{port}"
 
 
@@ -120,7 +143,7 @@ def stop_proxy(service_name):
     Args:
         service_name (str): Name of the Cloud Run service.
     """
-    proxies = load_proxies()
+    proxies = clean_proxy_list()
 
     if service_name not in proxies:
         print(f"No proxy found for service: {service_name}")
@@ -139,20 +162,99 @@ def stop_proxy(service_name):
     
     list_proxies()
 
+def stop_all_proxies():
+    """
+    Stops all running gcloud proxies.
+    """
+    proxies = clean_proxy_list()
+
+    for service_name, info in proxies.items():
+        pid = info["pid"]
+        try:
+            os.kill(pid, signal.SIGTERM)
+            print(f"Proxy for {service_name} stopped.")
+        except ProcessLookupError:
+            print(f"No process found with PID: {pid}")
+        except Exception as e:
+            print(f"Error stopping proxy for {service_name}: {e}")
+    
+    save_proxies({})
+
+    list_proxies()
+
+def list_cloud_run_services(project, region):
+    """
+    Lists all Cloud Run services the user has access to in a specific project and region.
+
+    Args:
+        project (str): The GCP project ID.
+        region (str): The region of the Cloud Run services.
+    """
+
+        # point or star?
+    with console.status("[bold orange]Listing Cloud Run Services[/bold orange]", spinner="star") as status:
+        try:
+            result = subprocess.run(
+                ["gcloud", "run", "services", "list", "--project", project, "--region", region, "--format=json"],
+                stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=30
+            )
+            if result.returncode != 0:
+                status.stop()
+                console.print(f"[bold red]ERROR: Unable to list Cloud Run services: {result.stderr.decode()}[/bold red]")
+                return
+
+            services = json.loads(result.stdout.decode())
+            if not services:
+                status.stop()
+                console.print("[bold red]No Cloud Run services found.[/bold red]")
+                return
+
+            proxies = clean_proxy_list()
+            status.stop()
+
+            table = Table(title="VAC Cloud Run Services")
+            table.add_column("Service Name")
+            table.add_column("Region")
+            table.add_column("URL")
+            table.add_column("Proxied")
+            table.add_column("Port")
+            
+            for service in services:
+                service_name = service['metadata']['name']
+                service_url = service['status']['url']
+                if service_name in proxies:
+                    proxied = "Yes"
+                    proxy_port = proxies[service_name]['port']
+                else:
+                    proxied = "No"
+                    proxy_port = "-"
+                table.add_row(service_name, region, service_url, proxied, str(proxy_port))
+
+            console.print(table)
+        except Exception as e:
+            status.stop()
+            console.print(f"[bold red]ERROR: An unexpected error occurred: {e}[/bold red]")
+
 def list_proxies():
     """
     Lists all running proxies.
     """
-    proxies = load_proxies()
+    print("Listing Proxies...")
+    proxies = clean_proxy_list()
     if not proxies:
         print("No proxies currently running.")
     else:
-        print("=== VAC Proxies Running ===")
+        table = Table(title="VAC Proxies")
+        table.add_column("VAC")
+        table.add_column("Port")
+        table.add_column("PID")
+        table.add_column("URL")
+        
         for service_name, info in proxies.items():
             url = f"http://127.0.0.1:{info['port']}"
-            hyperlink = create_hyperlink(url, url)
-            print(f"- VAC: {service_name} Port: {info['port']} PID: {info['pid']} URL: {hyperlink}")
-
+            table.add_row(service_name, str(info['port']), str(info['pid']), url)
+        
+        console.print(table)
 
 def setup_proxy_subparser(subparsers):
     """
@@ -161,13 +263,12 @@ def setup_proxy_subparser(subparsers):
     Args:
         subparsers: The subparsers object from argparse.ArgumentParser().
     """
+    
     proxy_parser = subparsers.add_parser('proxy', help='Set up or stop a proxy to the Cloud Run service.')
     proxy_subparsers = proxy_parser.add_subparsers(dest='proxy_command', required=True)
 
     start_parser = proxy_subparsers.add_parser('start', help='Start the proxy to the Cloud Run service.')
     start_parser.add_argument('service_name', help='Name of the Cloud Run service.')
-    start_parser.add_argument('--region', default='europe-west1', help='Region of the Cloud Run service.')
-    start_parser.add_argument('--project', default='multivac-internal-dev', help='GCP project of the Cloud Run service.')
     start_parser.add_argument('--port', type=int, help='Port to run the proxy on. Auto-assigns if not provided.')
     start_parser.set_defaults(func=lambda args: start_proxy(args.service_name, args.region, args.project, args.port))
 
@@ -178,4 +279,10 @@ def setup_proxy_subparser(subparsers):
     list_parser = proxy_subparsers.add_parser('list', help='List all running proxies.')
     list_parser.set_defaults(func=lambda args: list_proxies())
 
+    stop_all_parser = proxy_subparsers.add_parser('stop-all', help='Stop all running proxies.')
+    stop_all_parser.set_defaults(func=lambda args: stop_all_proxies())
+
+    list_services_parser = proxy_subparsers.add_parser('list-vacs', help='List all Cloud Run VAC services.')
+    list_services_parser.set_defaults(func=lambda args: list_cloud_run_services(args.project, args.region))
+
 

sunholo-0.60.3/sunholo/cli/sun_rich.py

@@ -0,0 +1,3 @@
+from rich.console import Console
+
+console = Console()

{sunholo-0.60.1 → sunholo-0.60.3}/sunholo/components/retriever.py

@@ -48,7 +48,8 @@ def pick_retriever(vector_name, embeddings=None):
                 log.info(f"Found vectorstore {vectorstore}")
                 if embeddings is None:
                     embeddings = get_embeddings(vector_name)
-                vectorstore = pick_vectorstore(vectorstore, vector_name=vector_name, embeddings=embeddings)
+                read_only = value.get('readonly')
+                vectorstore = pick_vectorstore(vectorstore, vector_name=vector_name, embeddings=embeddings, read_only=read_only)
                 k_override = value.get('k', 3)
                 vs_retriever = vectorstore.as_retriever(search_kwargs=dict(k=k_override))
                 retriever_list.append(vs_retriever)