jarviscore-framework 0.1.1__py3-none-any.whl → 0.2.1__py3-none-any.whl

tests/test_05_integration.py

@@ -0,0 +1,1301 @@
+"""
+Test 5: Integration Test - Multiple LLM-Powered Agents in Mesh
+
+Demonstrates the COMPLETE developer experience with MULTIPLE agents,
+all being FULL MESH PARTICIPANTS that can both SEND and RECEIVE.
+
+KEY CONCEPTS:
+    1. Every agent is LLM-powered
+    2. Every agent can SEND requests (via ask_peer)
+    3. Every agent can RECEIVE requests (via run() loop)
+    4. Every agent has get_tools() with peer tools
+    5. The role defines what they're GOOD at, not communication direction
+
+REAL-WORLD SCENARIO:
+    - Analyst: Good at analysis, but might ask researcher for data
+    - Researcher: Good at research, but might ask analyst to interpret
+    - Assistant: Good at chat/search, coordinates between specialists
+
+    All three can talk to each other in any direction!
+
+FILE STRUCTURE:
+    project/
+    ├── main.py              # Entry file (mesh setup)
+    ├── agents/
+    │   ├── analyst.py       # LLM agent - good at analysis
+    │   ├── researcher.py    # LLM agent - good at research
+    │   └── assistant.py     # LLM agent - good at chat/search
+    └── ...
+"""
+import asyncio
+import sys
+import pytest
+sys.path.insert(0, '.')
+
+from jarviscore.core.agent import Agent
+from jarviscore.core.mesh import Mesh
+from jarviscore.p2p.peer_client import PeerClient
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# AGENTS - All follow the SAME pattern, different capabilities
+# ═══════════════════════════════════════════════════════════════════════════════
+
+class Analyst(Agent):
+    """
+    Analyst - Good at analysis, can also ask other peers.
+
+    This agent might receive an analysis request, realize it needs
+    more data, and ask the researcher for help.
+    """
+    role = "analyst"
+    capabilities = ["analysis", "synthesis", "reporting"]
+
+    def __init__(self, agent_id=None):
+        super().__init__(agent_id)
+        self.analyses_count = 0
+        self.requests_received = []
+        self.requests_sent = []
+
+    def analyze(self, data: str) -> dict:
+        """Core capability - analyze data."""
+        self.analyses_count += 1
+        return {
+            "response": f"Analysis #{self.analyses_count}: '{data}' shows positive trends",
+            "confidence": 0.87
+        }
+
+    def get_tools(self) -> list:
+        """Return tools for LLM - includes peer tools."""
+        tools = [
+            {
+                "name": "analyze",
+                "description": "Analyze data and return insights",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {"data": {"type": "string"}},
+                    "required": ["data"]
+                }
+            }
+        ]
+        if self.peers:
+            tools.extend(self.peers.as_tool().schema)
+        return tools
+
+    async def execute_tool(self, tool_name: str, args: dict) -> str:
+        """Execute tool - routes to local or peer tools."""
+        if self.peers and tool_name in self.peers.as_tool().tool_names:
+            self.requests_sent.append({"tool": tool_name, "args": args})
+            return await self.peers.as_tool().execute(tool_name, args)
+        if tool_name == "analyze":
+            return str(self.analyze(args.get("data", "")))
+        return f"Unknown: {tool_name}"
+
+    async def run(self):
+        """Listen and respond to requests."""
+        while not self.shutdown_requested:
+            msg = await self.peers.receive(timeout=0.5)
+            if msg is None:
+                continue
+            if msg.is_request:
+                self.requests_received.append(msg.data)
+                result = self.analyze(msg.data.get("query", ""))
+                await self.peers.respond(msg, result)
+
+    async def execute_task(self, task): return {}
+
+
+class Researcher(Agent):
+    """
+    Researcher - Good at research, can also ask other peers.
+
+    This agent might receive a research request, get results,
+    and ask the analyst to interpret them.
+    """
+    role = "researcher"
+    capabilities = ["research", "data_collection", "summarization"]
+
+    def __init__(self, agent_id=None):
+        super().__init__(agent_id)
+        self.research_count = 0
+        self.requests_received = []
+        self.requests_sent = []
+
+    def research(self, topic: str) -> dict:
+        """Core capability - research a topic."""
+        self.research_count += 1
+        return {
+            "response": f"Research #{self.research_count}: Found 5 papers on '{topic}'",
+            "sources": ["paper1.pdf", "paper2.pdf"]
+        }
+
+    def get_tools(self) -> list:
+        """Return tools for LLM - includes peer tools."""
+        tools = [
+            {
+                "name": "research",
+                "description": "Research a topic and find relevant sources",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {"topic": {"type": "string"}},
+                    "required": ["topic"]
+                }
+            }
+        ]
+        if self.peers:
+            tools.extend(self.peers.as_tool().schema)
+        return tools
+
+    async def execute_tool(self, tool_name: str, args: dict) -> str:
+        """Execute tool - routes to local or peer tools."""
+        if self.peers and tool_name in self.peers.as_tool().tool_names:
+            self.requests_sent.append({"tool": tool_name, "args": args})
+            return await self.peers.as_tool().execute(tool_name, args)
+        if tool_name == "research":
+            return str(self.research(args.get("topic", "")))
+        return f"Unknown: {tool_name}"
+
+    async def run(self):
+        """Listen and respond to requests."""
+        while not self.shutdown_requested:
+            msg = await self.peers.receive(timeout=0.5)
+            if msg is None:
+                continue
+            if msg.is_request:
+                self.requests_received.append(msg.data)
+                result = self.research(msg.data.get("query", ""))
+                await self.peers.respond(msg, result)
+
+    async def execute_task(self, task): return {}
+
+
+class Assistant(Agent):
+    """
+    Assistant - Good at chat/search, coordinates between specialists.
+
+    This agent might receive a complex request and delegate parts
+    to analyst and researcher, then combine the results.
+    """
+    role = "assistant"
+    capabilities = ["chat", "search", "coordination"]
+
+    def __init__(self, agent_id=None):
+        super().__init__(agent_id)
+        self.search_count = 0
+        self.requests_received = []
+        self.requests_sent = []
+
+    def search(self, query: str) -> str:
+        """Core capability - search the web."""
+        self.search_count += 1
+        return f"Search #{self.search_count}: Results for '{query}'"
+
+    def get_tools(self) -> list:
+        """Return tools for LLM - includes peer tools."""
+        tools = [
+            {
+                "name": "search",
+                "description": "Search the web for information",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {"query": {"type": "string"}},
+                    "required": ["query"]
+                }
+            }
+        ]
+        if self.peers:
+            tools.extend(self.peers.as_tool().schema)
+        return tools
+
+    async def execute_tool(self, tool_name: str, args: dict) -> str:
+        """Execute tool - routes to local or peer tools."""
+        if self.peers and tool_name in self.peers.as_tool().tool_names:
+            self.requests_sent.append({"tool": tool_name, "args": args})
+            return await self.peers.as_tool().execute(tool_name, args)
+        if tool_name == "search":
+            return self.search(args.get("query", ""))
+        return f"Unknown: {tool_name}"
+
+    async def run(self):
+        """Listen and respond to requests."""
+        while not self.shutdown_requested:
+            msg = await self.peers.receive(timeout=0.5)
+            if msg is None:
+                continue
+            if msg.is_request:
+                self.requests_received.append(msg.data)
+                result = {"response": self.search(msg.data.get("query", ""))}
+                await self.peers.respond(msg, result)
+
+    async def execute_task(self, task): return {}
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# FIXTURES
+# ═══════════════════════════════════════════════════════════════════════════════
+
+@pytest.fixture
+def mesh():
+    """Create a fresh mesh."""
+    return Mesh(mode="p2p")
+
+
+@pytest.fixture
+def wired_mesh(mesh):
+    """Create mesh with all three agents wired up."""
+    analyst = mesh.add(Analyst)
+    researcher = mesh.add(Researcher)
+    assistant = mesh.add(Assistant)
+
+    for agent in mesh.agents:
+        agent.peers = PeerClient(
+            coordinator=None,
+            agent_id=agent.agent_id,
+            agent_role=agent.role,
+            agent_registry=mesh._agent_registry,
+            node_id="local"
+        )
+
+    return mesh, analyst, researcher, assistant
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# TESTS
+# ═══════════════════════════════════════════════════════════════════════════════
+
+class TestMeshSetup:
+    """Tests for basic mesh setup with multiple agents."""
+
+    def test_all_agents_registered(self, wired_mesh):
+        """All agents should be registered in the mesh."""
+        mesh, analyst, researcher, assistant = wired_mesh
+        assert len(mesh.agents) == 3
+
+    def test_all_agents_have_peers(self, wired_mesh):
+        """All agents should have peer client injected."""
+        mesh, analyst, researcher, assistant = wired_mesh
+        for agent in [analyst, researcher, assistant]:
+            assert agent.peers is not None
+
+    def test_all_agents_see_each_other(self, wired_mesh):
+        """Each agent should see the other two in their peer list."""
+        mesh, analyst, researcher, assistant = wired_mesh
+
+        analyst_peers = [p["role"] for p in analyst.peers.list_peers()]
+        assert "researcher" in analyst_peers
+        assert "assistant" in analyst_peers
+
+        researcher_peers = [p["role"] for p in researcher.peers.list_peers()]
+        assert "analyst" in researcher_peers
+        assert "assistant" in researcher_peers
+
+        assistant_peers = [p["role"] for p in assistant.peers.list_peers()]
+        assert "analyst" in assistant_peers
+        assert "researcher" in assistant_peers
+
+
+class TestAllAgentsHavePeerTools:
+    """Tests that ALL agents have peer tools in their toolset."""
+
+    def test_analyst_has_peer_tools(self, wired_mesh):
+        """Analyst should have ask_peer, broadcast_update, list_peers."""
+        mesh, analyst, researcher, assistant = wired_mesh
+        tools = analyst.get_tools()
+        tool_names = [t["name"] for t in tools]
+
+        assert "analyze" in tool_names  # Local
+        assert "ask_peer" in tool_names  # Peer
+        assert "broadcast_update" in tool_names
+        assert "list_peers" in tool_names
+
+    def test_researcher_has_peer_tools(self, wired_mesh):
+        """Researcher should have ask_peer, broadcast_update, list_peers."""
+        mesh, analyst, researcher, assistant = wired_mesh
+        tools = researcher.get_tools()
+        tool_names = [t["name"] for t in tools]
+
+        assert "research" in tool_names  # Local
+        assert "ask_peer" in tool_names  # Peer
+        assert "broadcast_update" in tool_names
+        assert "list_peers" in tool_names
+
+    def test_assistant_has_peer_tools(self, wired_mesh):
+        """Assistant should have ask_peer, broadcast_update, list_peers."""
+        mesh, analyst, researcher, assistant = wired_mesh
+        tools = assistant.get_tools()
+        tool_names = [t["name"] for t in tools]
+
+        assert "search" in tool_names  # Local
+        assert "ask_peer" in tool_names  # Peer
+        assert "broadcast_update" in tool_names
+        assert "list_peers" in tool_names
+
+
+class TestBidirectionalCommunication:
+    """Tests that prove ANY agent can talk to ANY other agent."""
+
+    @pytest.mark.asyncio
+    async def test_analyst_asks_researcher(self, wired_mesh):
+        """Analyst can ask researcher for data."""
+        mesh, analyst, researcher, assistant = wired_mesh
+
+        researcher_task = asyncio.create_task(researcher.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            result = await analyst.execute_tool("ask_peer", {
+                "role": "researcher",
+                "question": "Find papers on market trends"
+            })
+            assert "Research" in result
+            assert len(analyst.requests_sent) == 1
+
+        finally:
+            researcher.request_shutdown()
+            researcher_task.cancel()
+            try: await researcher_task
+            except asyncio.CancelledError: pass
+
+    @pytest.mark.asyncio
+    async def test_researcher_asks_analyst(self, wired_mesh):
+        """Researcher can ask analyst to interpret data."""
+        mesh, analyst, researcher, assistant = wired_mesh
+
+        analyst_task = asyncio.create_task(analyst.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            result = await researcher.execute_tool("ask_peer", {
+                "role": "analyst",
+                "question": "Interpret these research findings"
+            })
+            assert "Analysis" in result
+            assert len(researcher.requests_sent) == 1
+
+        finally:
+            analyst.request_shutdown()
+            analyst_task.cancel()
+            try: await analyst_task
+            except asyncio.CancelledError: pass
+
+    @pytest.mark.asyncio
+    async def test_assistant_coordinates_both(self, wired_mesh):
+        """Assistant can ask both analyst and researcher."""
+        mesh, analyst, researcher, assistant = wired_mesh
+
+        analyst_task = asyncio.create_task(analyst.run())
+        researcher_task = asyncio.create_task(researcher.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            # Ask researcher
+            r1 = await assistant.execute_tool("ask_peer", {
+                "role": "researcher",
+                "question": "Research AI trends"
+            })
+            assert "Research" in r1
+
+            # Ask analyst
+            r2 = await assistant.execute_tool("ask_peer", {
+                "role": "analyst",
+                "question": "Analyze the findings"
+            })
+            assert "Analysis" in r2
+
+            assert len(assistant.requests_sent) == 2
+
+        finally:
+            analyst.request_shutdown()
+            researcher.request_shutdown()
+            analyst_task.cancel()
+            researcher_task.cancel()
+            for t in [analyst_task, researcher_task]:
+                try: await t
+                except asyncio.CancelledError: pass
+
+
+class TestMultiAgentScenario:
+    """Tests for realistic multi-agent scenarios."""
+
+    @pytest.mark.asyncio
+    async def test_chain_of_requests(self, wired_mesh):
+        """
+        Test a chain: Assistant → Researcher → (gets data) → Assistant asks Analyst
+
+        This proves complex multi-agent workflows work.
+        """
+        mesh, analyst, researcher, assistant = wired_mesh
+
+        analyst_task = asyncio.create_task(analyst.run())
+        researcher_task = asyncio.create_task(researcher.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            # Step 1: Assistant asks researcher
+            research_result = await assistant.execute_tool("ask_peer", {
+                "role": "researcher",
+                "question": "Find data on Q4 sales"
+            })
+            assert "Research" in research_result
+
+            # Step 2: Assistant asks analyst to interpret
+            analysis_result = await assistant.execute_tool("ask_peer", {
+                "role": "analyst",
+                "question": f"Interpret: {research_result}"
+            })
+            assert "Analysis" in analysis_result
+
+            # Step 3: Assistant broadcasts completion
+            broadcast_result = await assistant.execute_tool("broadcast_update", {
+                "message": "Research and analysis complete!"
+            })
+            assert "Broadcast" in broadcast_result
+
+        finally:
+            analyst.request_shutdown()
+            researcher.request_shutdown()
+            for t in [analyst_task, researcher_task]:
+                t.cancel()
+                try: await t
+                except asyncio.CancelledError: pass
+
+    @pytest.mark.asyncio
+    async def test_all_agents_can_receive_while_sending(self, wired_mesh):
+        """
+        All agents running their run() loops while also sending.
+
+        This proves true bidirectional communication.
+        """
+        mesh, analyst, researcher, assistant = wired_mesh
+
+        # Start all run loops
+        analyst_task = asyncio.create_task(analyst.run())
+        researcher_task = asyncio.create_task(researcher.run())
+        assistant_task = asyncio.create_task(assistant.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            # Analyst sends to researcher
+            r1 = await analyst.execute_tool("ask_peer", {
+                "role": "researcher", "question": "Get data"
+            })
+            assert "Research" in r1
+
+            # Researcher sends to assistant
+            r2 = await researcher.execute_tool("ask_peer", {
+                "role": "assistant", "question": "Search for more"
+            })
+            assert "Search" in r2
+
+            # Assistant sends to analyst
+            r3 = await assistant.execute_tool("ask_peer", {
+                "role": "analyst", "question": "Analyze this"
+            })
+            assert "Analysis" in r3
+
+            # Verify all received requests
+            assert len(researcher.requests_received) >= 1  # From analyst
+            assert len(assistant.requests_received) >= 1   # From researcher
+            assert len(analyst.requests_received) >= 1     # From assistant
+
+        finally:
+            for agent in [analyst, researcher, assistant]:
+                agent.request_shutdown()
+            for t in [analyst_task, researcher_task, assistant_task]:
+                t.cancel()
+                try: await t
+                except asyncio.CancelledError: pass
+
+
+class TestLLMToolDispatch:
+    """Tests simulating LLM tool dispatch patterns."""
+
+    @pytest.mark.asyncio
+    async def test_llm_decides_to_ask_peer(self, wired_mesh):
+        """
+        Simulate LLM deciding to use ask_peer tool.
+
+        This is what happens in real code:
+        1. LLM receives request
+        2. LLM sees tools including ask_peer
+        3. LLM decides to delegate to specialist
+        4. Tool is executed
+        5. Result returned to LLM
+        """
+        mesh, analyst, researcher, assistant = wired_mesh
+
+        analyst_task = asyncio.create_task(analyst.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            # Step 1: Get tools (what LLM sees)
+            tools = assistant.get_tools()
+            tool_names = [t["name"] for t in tools]
+            assert "ask_peer" in tool_names
+
+            # Step 2: Simulate LLM decision
+            llm_decision = {
+                "tool": "ask_peer",
+                "args": {"role": "analyst", "question": "Analyze data"}
+            }
+
+            # Step 3: Execute
+            result = await assistant.execute_tool(
+                llm_decision["tool"],
+                llm_decision["args"]
+            )
+
+            # Step 4: Result is string for LLM
+            assert isinstance(result, str)
+
+        finally:
+            analyst.request_shutdown()
+            analyst_task.cancel()
+            try: await analyst_task
+            except asyncio.CancelledError: pass
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# LLM SIMULATION - Tests the COMPLETE LLM tool-use loop
+# ═══════════════════════════════════════════════════════════════════════════════
+
+class MockLLM:
+    """
+    Simulates LLM behavior for testing the complete tool-use flow.
+
+    In real-world:
+        - LLM receives: system prompt + tools + user message
+        - LLM decides: use a tool OR respond directly
+        - LLM returns: tool_call OR text response
+
+    This mock makes deterministic decisions based on keywords:
+        - "analyze" → delegate to analyst (unless I AM the analyst)
+        - "research" → delegate to researcher (unless I AM the researcher)
+        - "search" → use local search tool
+        - otherwise → respond directly
+    """
+
+    def __init__(self, agent_role: str):
+        self.agent_role = agent_role
+        self.calls = []  # Track all LLM calls for verification
+
+    def chat(self, messages: list, tools: list) -> dict:
+        """
+        Simulate LLM chat completion.
+
+        Returns either:
+            {"type": "tool_use", "tool": "name", "args": {...}}
+            {"type": "text", "content": "response"}
+        """
+        self.calls.append({"messages": messages, "tools": tools})
+
+        # Get the last user message
+        user_msg = ""
+        for msg in reversed(messages):
+            if msg.get("role") == "user":
+                user_msg = msg.get("content", "").lower()
+                break
+
+        # Get available tool names
+        tool_names = [t["name"] for t in tools]
+
+        # Decision logic (simulates LLM reasoning)
+        # KEY: If I AM the specialist, I process locally. Otherwise delegate.
+
+        if "analyze" in user_msg:
+            if self.agent_role == "analyst":
+                # I AM the analyst - process locally
+                return {
+                    "type": "text",
+                    "content": f"[analyst] Analysis complete: '{user_msg}' shows positive trends with 87% confidence"
+                }
+            elif "ask_peer" in tool_names:
+                # Delegate to analyst
+                return {
+                    "type": "tool_use",
+                    "tool": "ask_peer",
+                    "args": {"role": "analyst", "question": user_msg}
+                }
+
+        elif "research" in user_msg:
+            if self.agent_role == "researcher":
+                # I AM the researcher - process locally
+                return {
+                    "type": "text",
+                    "content": f"[researcher] Research complete: Found 5 papers on '{user_msg}'"
+                }
+            elif "ask_peer" in tool_names:
+                # Delegate to researcher
+                return {
+                    "type": "tool_use",
+                    "tool": "ask_peer",
+                    "args": {"role": "researcher", "question": user_msg}
+                }
+
+        elif "search" in user_msg and "search" in tool_names:
+            # LLM decides to use local search tool
+            return {
+                "type": "tool_use",
+                "tool": "search",
+                "args": {"query": user_msg}
+            }
+
+        elif "calculate" in user_msg and "calculate" in tool_names:
+            # LLM decides to use local calculate tool
+            expr = user_msg.split("calculate")[-1].strip()
+            return {
+                "type": "tool_use",
+                "tool": "calculate",
+                "args": {"expression": expr or "1+1"}
+            }
+
+        # Default: respond directly
+        return {
+            "type": "text",
+            "content": f"[{self.agent_role}] I can help with that: {user_msg}"
+        }
+
+    def incorporate_tool_result(self, tool_name: str, result: str) -> str:
+        """
+        Simulate LLM incorporating tool result into final response.
+
+        In real-world, this would be another LLM call with the tool result.
+        """
+        return f"Based on the {tool_name} result: {result}"
+
+
+class LLMPoweredAgent(Agent):
+    """
+    Agent with ACTUAL LLM integration (mocked for testing).
+
+    This is what a real-world agent looks like:
+        1. Has a MockLLM (would be real LLM in production)
+        2. chat() method that drives the LLM loop
+        3. Handles tool calls and incorporates results
+    """
+    role = "llm_agent"
+    capabilities = ["chat", "delegate"]
+
+    def __init__(self, agent_id=None, role_name="llm_agent"):
+        super().__init__(agent_id)
+        self.role = role_name
+        self.llm = MockLLM(role_name)
+        self.conversation_history = []
+        self.tool_calls_made = []
+
+    def get_tools(self) -> list:
+        """Return tools for LLM."""
+        tools = [
+            {
+                "name": "search",
+                "description": "Search for information",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {"query": {"type": "string"}},
+                    "required": ["query"]
+                }
+            },
+            {
+                "name": "calculate",
+                "description": "Calculate a math expression",
+                "input_schema": {
+                    "type": "object",
+                    "properties": {"expression": {"type": "string"}},
+                    "required": ["expression"]
+                }
+            }
+        ]
+        if self.peers:
+            tools.extend(self.peers.as_tool().schema)
+        return tools
+
+    async def execute_tool(self, tool_name: str, args: dict) -> str:
+        """Execute a tool."""
+        self.tool_calls_made.append({"tool": tool_name, "args": args})
+
+        if self.peers and tool_name in self.peers.as_tool().tool_names:
+            return await self.peers.as_tool().execute(tool_name, args)
+        if tool_name == "search":
+            return f"Search results for: {args.get('query', '')}"
+        if tool_name == "calculate":
+            try:
+                return f"Result: {eval(args.get('expression', '0'))}"
+            except:
+                return "Error in calculation"
+        return f"Unknown tool: {tool_name}"
+
+    async def chat(self, user_message: str) -> str:
+        """
+        Complete LLM chat loop with tool use.
+
+        This is THE KEY METHOD that shows real-world LLM tool use:
+        1. Add user message to history
+        2. Call LLM with messages + tools
+        3. If LLM returns tool_use → execute tool → call LLM again with result
+        4. If LLM returns text → return as final response
+        """
+        # Step 1: Add user message
+        self.conversation_history.append({
+            "role": "user",
+            "content": user_message
+        })
+
+        # Step 2: Get tools and call LLM
+        tools = self.get_tools()
+        llm_response = self.llm.chat(self.conversation_history, tools)
+
+        # Step 3: Handle tool use (may loop multiple times)
+        max_iterations = 5  # Prevent infinite loops
+        iteration = 0
+
+        while llm_response["type"] == "tool_use" and iteration < max_iterations:
+            iteration += 1
+
+            # Execute the tool
+            tool_name = llm_response["tool"]
+            tool_args = llm_response["args"]
+            tool_result = await self.execute_tool(tool_name, tool_args)
+
+            # Add tool call and result to history
+            self.conversation_history.append({
+                "role": "assistant",
+                "content": f"[Tool: {tool_name}] {tool_args}"
+            })
+            self.conversation_history.append({
+                "role": "tool",
+                "content": tool_result
+            })
+
+            # LLM incorporates result (simulated as another call)
+            final_response = self.llm.incorporate_tool_result(tool_name, tool_result)
+            llm_response = {"type": "text", "content": final_response}
+
+        # Step 4: Return final response
+        final = llm_response["content"]
+        self.conversation_history.append({
+            "role": "assistant",
+            "content": final
+        })
+
+        return final
+
+    async def run(self):
+        """Listen for incoming requests."""
+        while not self.shutdown_requested:
+            msg = await self.peers.receive(timeout=0.5)
+            if msg is None:
+                continue
+            if msg.is_request:
+                # Process with LLM
+                response = await self.chat(msg.data.get("query", ""))
+                await self.peers.respond(msg, {"response": response})
+
+    async def execute_task(self, task): return {}
+
+
+class TestLLMToolUseLoop:
+    """
+    Tests the COMPLETE LLM tool-use loop.
+
+    These tests prove that the ENTIRE flow works:
+        User message → LLM sees tools → LLM decides → Tool executes → Result back to LLM
+    """
+
+    @pytest.fixture
+    def llm_mesh(self):
+        """Create mesh with LLM-powered agents."""
+        mesh = Mesh(mode="p2p")
+
+        # Create agents with specific roles
+        assistant = LLMPoweredAgent(role_name="assistant")
+        analyst = LLMPoweredAgent(role_name="analyst")
+        researcher = LLMPoweredAgent(role_name="researcher")
+
+        # Manually register (registry stores lists of agents per role)
+        mesh._agent_registry["assistant"] = [assistant]
+        mesh._agent_registry["analyst"] = [analyst]
+        mesh._agent_registry["researcher"] = [researcher]
+        mesh.agents = [assistant, analyst, researcher]
+
+        # Wire up peers
+        for agent in mesh.agents:
+            agent.peers = PeerClient(
+                coordinator=None,
+                agent_id=agent.agent_id,
+                agent_role=agent.role,
+                agent_registry=mesh._agent_registry,
+                node_id="local"
+            )
+
+        return mesh, assistant, analyst, researcher
+
+    @pytest.mark.asyncio
+    async def test_llm_uses_local_tool(self, llm_mesh):
+        """
+        LLM decides to use a LOCAL tool (search).
+
+        Flow: User asks to search → LLM sees search tool → LLM uses it → Result returned
+        """
+        mesh, assistant, analyst, researcher = llm_mesh
+
+        # User asks to search
+        response = await assistant.chat("Please search for Python tutorials")
+
+        # Verify LLM used the search tool
+        assert len(assistant.tool_calls_made) == 1
+        assert assistant.tool_calls_made[0]["tool"] == "search"
+        assert "search" in response.lower() or "result" in response.lower()
+
+    @pytest.mark.asyncio
+    async def test_llm_delegates_to_peer(self, llm_mesh):
+        """
+        LLM decides to delegate to a PEER (analyst).
+
+        Flow: User asks for analysis → LLM sees ask_peer tool → LLM delegates → Peer responds
+        """
+        mesh, assistant, analyst, researcher = llm_mesh
+
+        # Start analyst listening
+        analyst_task = asyncio.create_task(analyst.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            # User asks for analysis
+            response = await assistant.chat("Please analyze the sales data")
+
+            # Verify LLM delegated to analyst
+            assert len(assistant.tool_calls_made) >= 1
+            peer_calls = [c for c in assistant.tool_calls_made if c["tool"] == "ask_peer"]
+            assert len(peer_calls) == 1
+            assert peer_calls[0]["args"]["role"] == "analyst"
+
+            # Verify response mentions the delegation
+            assert "ask_peer" in response.lower() or "result" in response.lower()
+
+        finally:
+            analyst.request_shutdown()
+            analyst_task.cancel()
+            try: await analyst_task
+            except asyncio.CancelledError: pass
+
+    @pytest.mark.asyncio
+    async def test_llm_responds_directly_when_appropriate(self, llm_mesh):
+        """
+        LLM responds directly without using tools.
+
+        Flow: User says hello → LLM doesn't need tools → LLM responds directly
+        """
+        mesh, assistant, analyst, researcher = llm_mesh
+
+        # User sends simple greeting
+        response = await assistant.chat("Hello, how are you?")
+
+        # Verify LLM did NOT use any tools
+        assert len(assistant.tool_calls_made) == 0
+
+        # Verify response is direct
+        assert "assistant" in response.lower() or "help" in response.lower()
+
+    @pytest.mark.asyncio
+    async def test_llm_sees_correct_tools(self, llm_mesh):
+        """
+        Verify LLM receives the correct tools including peer tools.
+        """
+        mesh, assistant, analyst, researcher = llm_mesh
+
+        tools = assistant.get_tools()
+        tool_names = [t["name"] for t in tools]
+
+        # Should have local tools
+        assert "search" in tool_names
+        assert "calculate" in tool_names
+
+        # Should have peer tools
+        assert "ask_peer" in tool_names
+        assert "broadcast_update" in tool_names
+        assert "list_peers" in tool_names
+
+    @pytest.mark.asyncio
+    async def test_llm_conversation_history_tracks_tool_use(self, llm_mesh):
+        """
+        Verify conversation history includes tool calls and results.
+        """
+        mesh, assistant, analyst, researcher = llm_mesh
+
+        # Make a request that uses a tool
+        await assistant.chat("Please search for AI news")
+
+        # Check conversation history
+        history = assistant.conversation_history
+
+        # Should have: user message, tool call, tool result, assistant response
+        assert len(history) >= 3
+
+        # Find tool-related entries
+        tool_entries = [h for h in history if "Tool:" in h.get("content", "")]
+        tool_results = [h for h in history if h.get("role") == "tool"]
+
+        assert len(tool_entries) >= 1  # Tool was called
+        assert len(tool_results) >= 1  # Result was recorded
+
+    @pytest.mark.asyncio
+    async def test_multi_agent_llm_conversation(self, llm_mesh):
+        """
+        Test a complex scenario where multiple agents use LLM to communicate.
+
+        Flow:
+        1. Assistant receives request needing research
+        2. Assistant's LLM delegates to researcher
+        3. Researcher processes and responds
+        4. Assistant's LLM incorporates result
+        """
+        mesh, assistant, analyst, researcher = llm_mesh
+
+        # Start both specialist agents
+        analyst_task = asyncio.create_task(analyst.run())
+        researcher_task = asyncio.create_task(researcher.run())
+        await asyncio.sleep(0.1)
+
+        try:
+            # User asks for research
+            response = await assistant.chat("Please research quantum computing")
+
+            # Verify delegation happened
+            peer_calls = [c for c in assistant.tool_calls_made if c["tool"] == "ask_peer"]
+            assert len(peer_calls) >= 1
+            assert peer_calls[0]["args"]["role"] == "researcher"
+
+        finally:
+            analyst.request_shutdown()
+            researcher.request_shutdown()
+            for t in [analyst_task, researcher_task]:
+                t.cancel()
+                try: await t
+                except asyncio.CancelledError: pass
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# FULL INTEGRATION - Complete scenario
+# ═══════════════════════════════════════════════════════════════════════════════
+
+async def test_full_integration():
+    """Complete integration test with all agents."""
+    print("\n" + "="*70)
+    print("FULL INTEGRATION: All Agents as Equal Mesh Participants")
+    print("="*70)
+
+    mesh = Mesh(mode="p2p")
+
+    analyst = mesh.add(Analyst)
+    researcher = mesh.add(Researcher)
+    assistant = mesh.add(Assistant)
+
+    for agent in mesh.agents:
+        agent.peers = PeerClient(
+            coordinator=None,
+            agent_id=agent.agent_id,
+            agent_role=agent.role,
+            agent_registry=mesh._agent_registry,
+            node_id="local"
+        )
+
+    # Start all listeners
+    tasks = [
+        asyncio.create_task(analyst.run()),
+        asyncio.create_task(researcher.run()),
+        asyncio.create_task(assistant.run())
+    ]
+    await asyncio.sleep(0.1)
+
+    print("\n[1] All agents see each other")
+    for agent in [analyst, researcher, assistant]:
+        peers = [p["role"] for p in agent.peers.list_peers()]
+        print(f"    {agent.role} sees: {peers}")
+
+    print("\n[2] All agents have peer tools")
+    for agent in [analyst, researcher, assistant]:
+        tools = [t["name"] for t in agent.get_tools()]
+        has_peer_tools = "ask_peer" in tools
+        print(f"    {agent.role}: {tools} (peer tools: {has_peer_tools})")
+
+    print("\n[3] Bidirectional communication")
+
+    # Analyst → Researcher
+    r = await analyst.execute_tool("ask_peer", {"role": "researcher", "question": "Get data"})
+    print(f"    Analyst → Researcher: {r[:40]}...")
+
+    # Researcher → Analyst
+    r = await researcher.execute_tool("ask_peer", {"role": "analyst", "question": "Interpret"})
+    print(f"    Researcher → Analyst: {r[:40]}...")
+
+    # Assistant → Both
+    r = await assistant.execute_tool("ask_peer", {"role": "analyst", "question": "Analyze"})
+    print(f"    Assistant → Analyst: {r[:40]}...")
+    r = await assistant.execute_tool("ask_peer", {"role": "researcher", "question": "Research"})
+    print(f"    Assistant → Researcher: {r[:40]}...")
+
+    print("\n[4] Request counts")
+    print(f"    Analyst received: {len(analyst.requests_received)}, sent: {len(analyst.requests_sent)}")
+    print(f"    Researcher received: {len(researcher.requests_received)}, sent: {len(researcher.requests_sent)}")
+    print(f"    Assistant received: {len(assistant.requests_received)}, sent: {len(assistant.requests_sent)}")
+
+    # Cleanup
+    for agent in [analyst, researcher, assistant]:
+        agent.request_shutdown()
+    for t in tasks:
+        t.cancel()
+        try: await t
+        except asyncio.CancelledError: pass
+
+    print("\n" + "="*70)
+    print("INTEGRATION TEST PASSED!")
+    print("="*70)
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# LLM TOOL-USE DEMO - Shows the complete flow with logging
+# ═══════════════════════════════════════════════════════════════════════════════
+
+class MockLLMWithLogging(MockLLM):
+    """MockLLM with detailed logging for demo purposes."""
+
+    def chat(self, messages: list, tools: list) -> dict:
+        """Chat with detailed logging."""
+        # Get the last user message
+        user_msg = ""
+        for msg in reversed(messages):
+            if msg.get("role") == "user":
+                user_msg = msg.get("content", "")
+                break
+
+        print(f"    │")
+        print(f"    ├─→ [LLM RECEIVES] Message: \"{user_msg}\"")
+        print(f"    │   Tools available: {[t['name'] for t in tools]}")
+
+        # Call parent to get decision
+        result = super().chat(messages, tools)
+
+        if result["type"] == "tool_use":
+            print(f"    │")
+            print(f"    ├─→ [LLM DECIDES] Use tool: {result['tool']}")
+            print(f"    │   Args: {result['args']}")
+        else:
+            print(f"    │")
+            print(f"    ├─→ [LLM DECIDES] Respond directly (no tool needed)")
+
+        return result
+
+    def incorporate_tool_result(self, tool_name: str, result: str) -> str:
+        """Incorporate with logging."""
+        print(f"    │")
+        print(f"    ├─→ [LLM RECEIVES RESULT] From {tool_name}:")
+        print(f"    │   Result: \"{result[:60]}...\"" if len(result) > 60 else f"    │   Result: \"{result}\"")
+
+        final = super().incorporate_tool_result(tool_name, result)
+
+        print(f"    │")
+        print(f"    └─→ [LLM RESPONDS] \"{final[:60]}...\"" if len(final) > 60 else f"    └─→ [LLM RESPONDS] \"{final}\"")
+
+        return final
+
+
+class LLMPoweredAgentWithLogging(LLMPoweredAgent):
+    """LLM-powered agent with detailed logging."""
+
+    def __init__(self, agent_id=None, role_name="llm_agent"):
+        super().__init__(agent_id, role_name)
+        self.llm = MockLLMWithLogging(role_name)  # Use logging version
+
+    async def execute_tool(self, tool_name: str, args: dict) -> str:
+        """Execute tool with logging."""
+        print(f"    │")
+        print(f"    ├─→ [EXECUTE TOOL] {tool_name}")
+
+        result = await super().execute_tool(tool_name, args)
+
+        if tool_name == "ask_peer":
+            print(f"    │   Sent to peer: {args.get('role')}")
+            print(f"    │   Peer responded: \"{result[:50]}...\"" if len(result) > 50 else f"    │   Peer responded: \"{result}\"")
+
+        return result
+
+
+async def demo_llm_tool_use():
+    """Demo the complete LLM tool-use flow with detailed logging."""
+    print("\n" + "="*70)
+    print("LLM TOOL-USE FLOW DEMO")
+    print("="*70)
+
+    print("""
+    This demo shows EXACTLY what happens when an LLM uses tools:
+
+    ┌─────────────────────────────────────────────────────────────────┐
+    │  User Message                                                   │
+    │       ↓                                                         │
+    │  LLM sees tools (local + peer)                                  │
+    │       ↓                                                         │
+    │  LLM decides: use tool OR respond directly                      │
+    │       ↓                                                         │
+    │  If tool: execute → get result → LLM incorporates               │
+    │       ↓                                                         │
+    │  Final response to user                                         │
+    └─────────────────────────────────────────────────────────────────┘
+    """)
+
+    # Setup mesh with logging agents
+    mesh = Mesh(mode="p2p")
+
+    assistant = LLMPoweredAgentWithLogging(role_name="assistant")
+    analyst = LLMPoweredAgentWithLogging(role_name="analyst")
+    researcher = LLMPoweredAgentWithLogging(role_name="researcher")
+
+    mesh._agent_registry["assistant"] = [assistant]
+    mesh._agent_registry["analyst"] = [analyst]
+    mesh._agent_registry["researcher"] = [researcher]
+    mesh.agents = [assistant, analyst, researcher]
+
+    for agent in mesh.agents:
+        agent.peers = PeerClient(
+            coordinator=None,
+            agent_id=agent.agent_id,
+            agent_role=agent.role,
+            agent_registry=mesh._agent_registry,
+            node_id="local"
+        )
+
+    # Start listeners
+    analyst_task = asyncio.create_task(analyst.run())
+    researcher_task = asyncio.create_task(researcher.run())
+    await asyncio.sleep(0.1)
+
+    # ─────────────────────────────────────────────────────────────────
+    # SCENARIO 1: LLM uses LOCAL tool
+    # ─────────────────────────────────────────────────────────────────
+    print("\n" + "─"*70)
+    print("SCENARIO 1: User asks to SEARCH (LLM uses LOCAL tool)")
+    print("─"*70)
+    print(f"\n[USER] → Assistant: \"Please search for Python tutorials\"")
+    print(f"\n[ASSISTANT LLM FLOW]")
+
+    response = await assistant.chat("Please search for Python tutorials")
+
+    print(f"\n[FINAL RESPONSE] → User: \"{response}\"")
+    print(f"\n✓ LLM decided to use LOCAL tool (search)")
+
+    # Reset for next scenario
+    assistant.tool_calls_made = []
+    assistant.conversation_history = []
+
+    # ─────────────────────────────────────────────────────────────────
+    # SCENARIO 2: LLM delegates to PEER
+    # ─────────────────────────────────────────────────────────────────
+    print("\n" + "─"*70)
+    print("SCENARIO 2: User asks for ANALYSIS (LLM delegates to PEER)")
+    print("─"*70)
+    print(f"\n[USER] → Assistant: \"Please analyze the Q4 sales data\"")
+    print(f"\n[ASSISTANT LLM FLOW]")
+
+    response = await assistant.chat("Please analyze the Q4 sales data")
+
+    print(f"\n[FINAL RESPONSE] → User: \"{response}\"")
+    print(f"\n✓ LLM decided to delegate to PEER (analyst) via ask_peer tool")
+
+    # Reset for next scenario
+    assistant.tool_calls_made = []
+    assistant.conversation_history = []
+
+    # ─────────────────────────────────────────────────────────────────
+    # SCENARIO 3: LLM responds directly
+    # ─────────────────────────────────────────────────────────────────
+    print("\n" + "─"*70)
+    print("SCENARIO 3: User says HELLO (LLM responds directly, no tools)")
+    print("─"*70)
+    print(f"\n[USER] → Assistant: \"Hello, how are you?\"")
+    print(f"\n[ASSISTANT LLM FLOW]")
+
+    response = await assistant.chat("Hello, how are you?")
+
+    print(f"\n[FINAL RESPONSE] → User: \"{response}\"")
+    print(f"\n✓ LLM decided NO tool needed, responded directly")
+
+    # ─────────────────────────────────────────────────────────────────
+    # SCENARIO 4: Multi-step - research then analyze
+    # ─────────────────────────────────────────────────────────────────
+    print("\n" + "─"*70)
+    print("SCENARIO 4: User asks for RESEARCH (LLM delegates to researcher)")
+    print("─"*70)
+    print(f"\n[USER] → Assistant: \"Please research AI trends\"")
+    print(f"\n[ASSISTANT LLM FLOW]")
+
+    response = await assistant.chat("Please research AI trends")
+
+    print(f"\n[FINAL RESPONSE] → User: \"{response}\"")
+    print(f"\n✓ LLM decided to delegate to PEER (researcher) via ask_peer tool")
+
+    # Cleanup
+    analyst.request_shutdown()
+    researcher.request_shutdown()
+    for t in [analyst_task, researcher_task]:
+        t.cancel()
+        try: await t
+        except asyncio.CancelledError: pass
+
+    print("\n" + "="*70)
+    print("LLM TOOL-USE DEMO COMPLETE!")
+    print("="*70)
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# MANUAL RUN
+# ═══════════════════════════════════════════════════════════════════════════════
+
+if __name__ == "__main__":
+    # Run mesh integration first
+    asyncio.run(test_full_integration())
+
+    print("""
+═══════════════════════════════════════════════════════════════════════════════
+KEY INSIGHT: ALL agents are EQUAL mesh participants
+═══════════════════════════════════════════════════════════════════════════════
+
+  Every agent (Analyst, Researcher, Assistant):
+  ├── Has an LLM for reasoning
+  ├── Has get_tools() with LOCAL + PEER tools
+  ├── Can SEND via ask_peer, broadcast_update
+  ├── Can RECEIVE via run() loop
+  └── The role just defines what they're GOOD at
+
+  Communication is bidirectional:
+  ├── Analyst ←→ Researcher
+  ├── Researcher ←→ Assistant
+  └── Assistant ←→ Analyst
+
+  This is the power of the mesh:
+  - No hierarchies
+  - No "sender" vs "receiver" types
+  - Every agent is a full participant
+═══════════════════════════════════════════════════════════════════════════════
+""")
+
+    # Run LLM tool-use demo
+    asyncio.run(demo_llm_tool_use())
+
+    print("""
+═══════════════════════════════════════════════════════════════════════════════
+KEY INSIGHT: LLM TOOL-USE FLOW
+═══════════════════════════════════════════════════════════════════════════════
+
+  The LLM sees ALL tools (local + peer) and DECIDES:
+
+  ┌─────────────────────────────────────────────────────────────────────────┐
+  │  "search for X"     → LLM uses LOCAL tool (search)                      │
+  │  "analyze X"        → LLM uses PEER tool (ask_peer → analyst)           │
+  │  "research X"       → LLM uses PEER tool (ask_peer → researcher)        │
+  │  "hello"            → LLM responds DIRECTLY (no tool needed)            │
+  └─────────────────────────────────────────────────────────────────────────┘
+
+  The framework provides the tools.
+  The LLM decides WHEN and HOW to use them.
+  PeerTool makes other agents available as tools!
+═══════════════════════════════════════════════════════════════════════════════
+""")