@macaw-ai/client 0.0.0-stage → 0.9.11

package/payload/secureAI/examples/langchain/langchain_1b_multiuser.py

@@ -0,0 +1,262 @@
+#!/usr/bin/env python3
+"""
+langchain_1b_multiuser.py - Multi-user Agents with role-based policies
+
+Create different agent instances per user role, each with their own security policy.
+Use this when: Different users need different tool access levels.
+
+In this example:
+- alice (Analyst): Can only use calculator
+- bob (Manager): Can use calculator and weather
+- carol (Admin): Can use all tools including admin
+
+Prerequisites:
+    - MACAW SDK installed (pip install macaw-client macaw-adapters)
+    - OPENAI_API_KEY environment variable (for ChatOpenAI)
+
+Run:
+    export OPENAI_API_KEY=sk-...
+    python langchain_1b_multiuser.py
+"""
+
+import os
+
+from macaw_adapters.langchain import create_react_agent, AgentExecutor, cleanup
+
+from langchain_openai import ChatOpenAI
+from langchain.tools import Tool
+from langchain.prompts import PromptTemplate
+
+
+# Define tools
+def calculator(expression: str) -> str:
+    """Evaluate a math expression."""
+    try:
+        result = eval(expression)
+        return f"Result: {result}"
+    except Exception as e:
+        return f"Error: {e}"
+
+
+def get_weather(city: str) -> str:
+    """Get weather for a city (mock)."""
+    weather_data = {
+        "new york": "72°F, Sunny",
+        "london": "58°F, Cloudy",
+        "tokyo": "68°F, Clear"
+    }
+    return weather_data.get(city.lower(), f"Weather for {city}: 65°F, Partly cloudy")
+
+
+def admin_tool(command: str) -> str:
+    """Execute admin command."""
+    return f"Executed admin command: {command}"
+
+
+def file_reader(path: str) -> str:
+    """Read a file."""
+    # Mock - returns different content based on path
+    if "report" in path.lower():
+        return "Q4 Revenue: $1.2M, Growth: 15%"
+    elif "secret" in path.lower():
+        return "ACCESS DENIED - Classified content"
+    return f"Contents of {path}: [mock data]"
+
+
+# All available tools
+ALL_TOOLS = [
+    Tool(name="calculator", func=calculator, description="Math calculations"),
+    Tool(name="weather", func=get_weather, description="Get weather for a city"),
+    Tool(name="admin", func=admin_tool, description="Execute admin commands"),
+    Tool(name="file_reader", func=file_reader, description="Read files"),
+]
+
+
+# User role configurations with MAPL format policies
+USER_CONFIGS = {
+    "alice": {
+        "role": "Financial Analyst",
+        "security_policy": {
+            "resources": ["tool:calculator", "tool:file_reader"],
+            "denied_resources": ["tool:admin"],
+            "constraints": {
+                "parameters": {
+                    "tool:file_reader": {"input": ["*report*", "*public*"]}
+                },
+                "denied_parameters": {
+                    "tool:*": {"input": ["*password*", "*secret*"]}
+                }
+            }
+        },
+        "tests": [
+            ("What is 100 * 12?", "ALLOWED - calculator"),
+            ("Read the Q4 report", "ALLOWED - file_reader for reports"),
+            ("Run admin command: list users", "BLOCKED - admin tool"),
+            ("What's the weather in London?", "BLOCKED - weather not allowed"),
+        ]
+    },
+    "bob": {
+        "role": "Finance Manager",
+        "security_policy": {
+            "resources": ["tool:calculator", "tool:weather", "tool:file_reader"],
+            "denied_resources": ["tool:admin"],
+            "constraints": {
+                "parameters": {
+                    "tool:file_reader": {"input": ["*report*", "*forecast*", "*public*"]}
+                },
+                "denied_parameters": {
+                    "tool:*": {"input": ["*password*"]}
+                }
+            }
+        },
+        "tests": [
+            ("What is 250 + 750?", "ALLOWED - calculator"),
+            ("What's the weather in Tokyo?", "ALLOWED - weather"),
+            ("Read the Q4 report", "ALLOWED - file_reader"),
+            ("Run admin command: restart", "BLOCKED - admin tool"),
+        ]
+    },
+    "carol": {
+        "role": "IT Administrator",
+        "security_policy": {
+            "resources": ["tool:calculator", "tool:weather", "tool:file_reader", "tool:admin"],
+            # No denied_resources - full tool access
+            "constraints": {
+                "parameters": {
+                    "tool:file_reader": {"input": ["*"]}  # All files
+                }
+            }
+        },
+        "tests": [
+            ("What is 999 * 111?", "ALLOWED - calculator"),
+            ("What's the weather in New York?", "ALLOWED - weather"),
+            ("Run admin command: status check", "ALLOWED - admin access"),
+            ("Read the secret file", "ALLOWED - full file access"),
+        ]
+    }
+}
+
+
+# Simple ReAct prompt
+REACT_PROMPT = PromptTemplate.from_template("""Answer the following questions as best you can. You have access to the following tools:
+
+{tools}
+
+Use the following format:
+
+Question: the input question you must answer
+Thought: you should always think about what to do
+Action: the action to take, should be one of [{tool_names}]
+Action Input: the input to the action
+Observation: the result of the action
+... (this Thought/Action/Action Input/Observation can repeat N times)
+Thought: I now know the final answer
+Final Answer: the final answer to the original input question
+
+Begin!
+
+Question: {input}
+Thought:{agent_scratchpad}""")
+
+
+def create_user_executor(username: str, config: dict, llm) -> AgentExecutor:
+    """Create a security-scoped executor for a specific user."""
+    print(f"\n  Creating executor for {username} ({config['role']})")
+    print(f"  Policy: {config['security_policy'].get('resources', 'all')}")
+
+    agent = create_react_agent(
+        llm=llm,
+        tools=ALL_TOOLS,
+        prompt=REACT_PROMPT,
+        security_policy=config["security_policy"]
+    )
+
+    executor = AgentExecutor(
+        agent=agent,
+        tools=ALL_TOOLS,
+        security_policy=config["security_policy"],
+        verbose=False,
+        handle_parsing_errors=True,
+        max_iterations=3
+    )
+
+    return executor
+
+
+def test_user(username: str, config: dict, executor: AgentExecutor):
+    """Test a user's executor with their test cases."""
+    print(f"\n{'=' * 60}")
+    print(f"Testing {username.upper()} ({config['role']})")
+    print(f"Allowed tools: {config['security_policy'].get('resources', 'all')}")
+    print("=" * 60)
+
+    for query, expected in config["tests"]:
+        print(f"\n  Query: {query}")
+        print(f"  Expected: {expected}")
+
+        try:
+            result = executor.invoke({"input": query})
+            output = result.get("output", str(result))
+
+            # Check if it was blocked
+            if "denied" in output.lower() or "access denied" in output.lower():
+                print(f"  Result: BLOCKED - {output[:50]}...")
+            elif "cannot" in output.lower() or "don't have" in output.lower():
+                print(f"  Result: BLOCKED (agent refused) - {output[:50]}...")
+            else:
+                print(f"  Result: SUCCESS - {output[:60]}...")
+
+        except Exception as e:
+            error_msg = str(e).lower()
+            if any(word in error_msg for word in ["denied", "blocked", "policy", "not allowed"]):
+                print(f"  Result: CORRECTLY BLOCKED - {str(e)[:50]}...")
+            else:
+                print(f"  Result: ERROR - {str(e)[:50]}...")
+
+
+def main():
+    if not os.environ.get("OPENAI_API_KEY"):
+        print("Set OPENAI_API_KEY environment variable")
+        return
+
+    print("=" * 60)
+    print("Example 1b: Multi-user Agents (LangChain)")
+    print("=" * 60)
+    print("\nPattern: Separate executor per user with role-based policies")
+    print("Use when: Different users need different tool access levels")
+    print("\nUsers:")
+    print("  alice (Analyst): calculator, file_reader (reports only)")
+    print("  bob (Manager): calculator, weather, file_reader")
+    print("  carol (Admin): all tools, all files")
+
+    # Shared LLM
+    llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
+
+    # Create and test each user
+    for username, config in USER_CONFIGS.items():
+        executor = create_user_executor(username, config, llm)
+        test_user(username, config, executor)
+
+    # Cleanup
+    print("\n--- Cleanup ---")
+    cleanup()
+
+    print("\n" + "=" * 60)
+    print("Example complete!")
+    print("=" * 60)
+
+
+if __name__ == "__main__":
+    try:
+        main()
+    except Exception as e:
+        err = str(e)
+        print("\n" + "=" * 60)
+        if "OPENAI_API_KEY" in err or "api_key" in err.lower():
+            print("ERROR: OpenAI API key not configured")
+            print("Fix: export OPENAI_API_KEY=sk-...")
+        else:
+            print(f"ERROR: {e}")
+        print("=" * 60)
+        import sys
+        sys.exit(1)

package/payload/secureAI/examples/langchain/langchain_1c_orchestration.py

@@ -0,0 +1,332 @@
+#!/usr/bin/env python3
+"""
+langchain_1c_orchestration.py - Agent Orchestration with supervisor routing
+
+A supervisor agent routes tasks to specialized sub-agents, each with different
+security policies. Demonstrates policy composition where chains can only
+restrict, never relax permissions.
+
+Use this when: Building multi-agent workflows with different permission levels.
+
+In this example:
+- Supervisor: Routes requests to appropriate agent
+- Research Agent: Can use search, file_reader (public only)
+- Finance Agent: Can use calculator, file_reader (reports only)
+- Admin Agent: Can use admin tools (restricted access)
+
+Prerequisites:
+    - MACAW SDK installed (pip install macaw-client macaw-adapters)
+    - OPENAI_API_KEY environment variable (for ChatOpenAI)
+
+Run:
+    export OPENAI_API_KEY=sk-...
+    python langchain_1c_orchestration.py
+"""
+
+import os
+from typing import Dict, Any
+
+from macaw_adapters.langchain import create_react_agent, AgentExecutor, cleanup
+
+from langchain_openai import ChatOpenAI
+from langchain.tools import Tool
+from langchain.prompts import PromptTemplate
+
+
+# Define specialized tools for different agents
+def calculator(expression: str) -> str:
+    """Evaluate a math expression."""
+    try:
+        result = eval(expression)
+        return f"Result: {result}"
+    except Exception as e:
+        return f"Error: {e}"
+
+
+def search_web(query: str) -> str:
+    """Search the web (mock)."""
+    return f"Search results for '{query}': [Top 3 relevant articles about {query}]"
+
+
+def file_reader(path: str) -> str:
+    """Read a file."""
+    if "report" in path.lower():
+        return "Q4 Revenue: $1.2M, Growth: 15%, Expenses: $800K"
+    elif "public" in path.lower():
+        return "Public announcement: Company expanding to new markets"
+    elif "secret" in path.lower() or "confidential" in path.lower():
+        return "ACCESS DENIED - Classified content"
+    return f"Contents of {path}: [document data]"
+
+
+def admin_tool(command: str) -> str:
+    """Execute admin command."""
+    return f"Admin executed: {command}"
+
+
+def send_email(input_str: str) -> str:
+    """Send an email (mock). Input format: 'to@email.com: subject text'"""
+    # Parse simple format: "to: subject" or just treat whole thing as recipient
+    if ":" in input_str:
+        parts = input_str.split(":", 1)
+        to = parts[0].strip()
+        subject = parts[1].strip() if len(parts) > 1 else "No subject"
+    else:
+        to = input_str.strip()
+        subject = "Notification"
+    return f"Email sent to {to}: {subject}"
+
+
+# All available tools
+ALL_TOOLS = [
+    Tool(name="calculator", func=calculator, description="Math calculations"),
+    Tool(name="search", func=search_web, description="Search the web for information"),
+    Tool(name="file_reader", func=file_reader, description="Read files"),
+    Tool(name="admin", func=admin_tool, description="Execute admin commands"),
+    Tool(name="email", func=send_email, description="Send emails"),
+]
+
+
+# ReAct prompt template
+REACT_PROMPT = PromptTemplate.from_template("""Answer the following questions as best you can. You have access to the following tools:
+
+{tools}
+
+Use the following format:
+
+Question: the input question you must answer
+Thought: you should always think about what to do
+Action: the action to take, should be one of [{tool_names}]
+Action Input: the input to the action
+Observation: the result of the action
+... (this Thought/Action/Action Input/Observation can repeat N times)
+Thought: I now know the final answer
+Final Answer: the final answer to the original input question
+
+Begin!
+
+Question: {input}
+Thought:{agent_scratchpad}""")
+
+
+# Specialized agent configurations with MAPL format policies
+AGENT_CONFIGS = {
+    "research": {
+        "description": "Research Agent - web search and public documents",
+        "policy": {
+            "resources": ["tool:search", "tool:file_reader"],
+            "denied_resources": ["tool:admin", "tool:calculator", "tool:email"],
+            "constraints": {
+                "parameters": {
+                    "tool:file_reader": {"input": ["*public*"]}
+                },
+                "denied_parameters": {
+                    "tool:*": {"input": ["*password*", "*secret*", "*confidential*"]}
+                }
+            }
+        }
+    },
+    "finance": {
+        "description": "Finance Agent - calculations and financial reports",
+        "policy": {
+            "resources": ["tool:calculator", "tool:file_reader"],
+            "denied_resources": ["tool:admin", "tool:search", "tool:email"],
+            "constraints": {
+                "parameters": {
+                    "tool:file_reader": {"input": ["*report*", "*forecast*", "*budget*"]}
+                },
+                "denied_parameters": {
+                    "tool:*": {"input": ["*password*"]}
+                }
+            }
+        }
+    },
+    "admin": {
+        "description": "Admin Agent - system administration (restricted)",
+        "policy": {
+            "resources": ["tool:admin", "tool:email"],
+            "denied_resources": ["tool:calculator", "tool:search", "tool:file_reader"],
+            "constraints": {
+                "denied_parameters": {
+                    "tool:*": {"input": ["*delete*", "*drop*", "*truncate*"]}
+                }
+            }
+        }
+    }
+}
+
+
+def create_specialized_agent(name: str, config: Dict[str, Any], llm) -> AgentExecutor:
+    """Create a specialized agent with its security policy."""
+    agent = create_react_agent(
+        llm=llm,
+        tools=ALL_TOOLS,
+        prompt=REACT_PROMPT,
+        security_policy=config["policy"]
+    )
+
+    return AgentExecutor(
+        agent=agent,
+        tools=ALL_TOOLS,
+        security_policy=config["policy"],
+        verbose=False,
+        handle_parsing_errors=True,
+        max_iterations=3
+    )
+
+
+class SupervisorAgent:
+    """
+    Supervisor that routes tasks to appropriate specialized agents.
+
+    Key security concept: The supervisor can only delegate to agents
+    with equal or more restrictive policies. It cannot grant permissions
+    it doesn't have (policy intersection).
+    """
+
+    def __init__(self, llm):
+        self.llm = llm
+        self.agents: Dict[str, AgentExecutor] = {}
+
+        # Create specialized agents
+        print("\n  Creating specialized agents:")
+        for name, config in AGENT_CONFIGS.items():
+            print(f"    - {name}: {config['description']}")
+            self.agents[name] = create_specialized_agent(name, config, llm)
+
+    def route(self, query: str) -> str:
+        """Route query to appropriate agent based on content."""
+        query_lower = query.lower()
+
+        # Simple routing logic (in production, could use LLM for routing)
+        if any(word in query_lower for word in ["calculate", "math", "revenue", "profit", "cost", "budget"]):
+            return "finance"
+        elif any(word in query_lower for word in ["search", "find", "lookup", "research", "public"]):
+            return "research"
+        elif any(word in query_lower for word in ["admin", "restart", "status", "notify", "email"]):
+            return "admin"
+        else:
+            return "research"  # Default to research
+
+    def invoke(self, query: str) -> Dict[str, Any]:
+        """Route and execute query through appropriate agent."""
+        agent_name = self.route(query)
+        agent = self.agents[agent_name]
+
+        print(f"  Routing to: {agent_name}")
+        print(f"  Policy: {AGENT_CONFIGS[agent_name]['policy'].get('resources', 'all')}")
+
+        try:
+            result = agent.invoke({"input": query})
+            return {
+                "routed_to": agent_name,
+                "success": True,
+                "output": result.get("output", str(result))
+            }
+        except Exception as e:
+            error_msg = str(e).lower()
+            if any(word in error_msg for word in ["denied", "blocked", "policy", "not allowed"]):
+                return {
+                    "routed_to": agent_name,
+                    "success": False,
+                    "blocked": True,
+                    "output": f"Policy blocked: {str(e)[:80]}"
+                }
+            return {
+                "routed_to": agent_name,
+                "success": False,
+                "output": f"Error: {str(e)[:80]}"
+            }
+
+
+def main():
+    if not os.environ.get("OPENAI_API_KEY"):
+        print("Set OPENAI_API_KEY environment variable")
+        return
+
+    print("=" * 60)
+    print("Example 1c: Agent Orchestration (LangChain)")
+    print("=" * 60)
+    print("\nPattern: Supervisor routes to specialized sub-agents")
+    print("Use when: Multi-agent workflows with different permission levels")
+    print("\nArchitecture:")
+    print("  Supervisor --> Research Agent (search, public files)")
+    print("            --> Finance Agent (calculator, reports)")
+    print("            --> Admin Agent (admin tools, email)")
+
+    # Shared LLM
+    llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
+
+    # Create supervisor
+    print("\n--- Creating Supervisor Agent ---")
+    supervisor = SupervisorAgent(llm)
+
+    # Test cases demonstrating routing and policy enforcement
+    TEST_CASES = [
+        # Should route to finance, use calculator
+        ("Calculate the profit if revenue is 1200000 and expenses are 800000", "finance", True),
+
+        # Should route to research, use search
+        ("Search for public information about AI trends", "research", True),
+
+        # Should route to admin, use email
+        ("Send an email to team@company.com about the system status", "admin", True),
+
+        # Should route to finance, but try to access secret file (blocked - not in allowed patterns)
+        ("What is the profit in the secret internal file?", "finance", False),
+
+        # Should route to research, but research agent can't use calculator
+        ("Search for how to compute 100 * 50", "research", True),  # Research will search, not calculate
+    ]
+
+    print("\n" + "=" * 60)
+    print("Running test cases")
+    print("=" * 60)
+
+    for query, expected_agent, should_succeed in TEST_CASES:
+        print(f"\n  Query: {query}")
+        print(f"  Expected: routes to '{expected_agent}', {'succeeds' if should_succeed else 'blocked'}")
+
+        result = supervisor.invoke(query)
+
+        routed = result["routed_to"]
+        succeeded = result["success"]
+        output = result["output"][:60] + "..." if len(result.get("output", "")) > 60 else result.get("output", "")
+
+        # Check if routing was correct
+        route_correct = routed == expected_agent
+        outcome_correct = succeeded == should_succeed
+
+        status = "PASS" if (route_correct and outcome_correct) else "UNEXPECTED"
+
+        print(f"  Result: [{status}] Routed to '{routed}', {'succeeded' if succeeded else 'blocked/failed'}")
+        print(f"  Output: {output}")
+
+    # Cleanup
+    print("\n--- Cleanup ---")
+    cleanup()
+
+    print("\n" + "=" * 60)
+    print("Example complete!")
+    print("=" * 60)
+    print("\nKey concepts demonstrated:")
+    print("  - Supervisor pattern for agent orchestration")
+    print("  - Each sub-agent has isolated security policy")
+    print("  - Policy enforcement at tool execution time")
+    print("  - Routing based on task type")
+
+
+if __name__ == "__main__":
+    try:
+        main()
+    except Exception as e:
+        err = str(e)
+        print("\n" + "=" * 60)
+        if "OPENAI_API_KEY" in err or "api_key" in err.lower():
+            print("ERROR: OpenAI API key not configured")
+            print("Fix: export OPENAI_API_KEY=sk-...")
+        else:
+            print(f"ERROR: {e}")
+        print("=" * 60)
+        import sys
+        sys.exit(1)