agent0-sdk 0.31__py3-none-any.whl

agent0_sdk/core/indexer.py

@@ -0,0 +1,1754 @@
+"""
+Agent indexer for discovery and search functionality.
+
+ARCHITECTURAL PURPOSE:
+======================
+
+The indexer serves as the unified entry point for all discovery and search operations
+(agents AND feedback), not merely a thin wrapper around SubgraphClient. While currently
+it delegates most queries to the subgraph, it is designed to be the foundation for:
+
+1. SEMANTIC/VECTOR SEARCH: Future integration with embeddings and vector databases
+   for semantic search across agent descriptions, feedback text, and capabilities.
+
+2. HYBRID SEARCH: Combining subgraph queries (structured data) with vector similarity
+   (semantic understanding) for richer discovery experiences.
+
+3. LOCAL INDEXING: Optional local caching and indexing for offline-capable applications
+   or performance optimization.
+
+4. SEARCH OPTIMIZATION: Advanced filtering, ranking, and relevance scoring that goes
+   beyond simple subgraph queries.
+
+5. MULTI-SOURCE AGGREGATION: Combining data from subgraph, blockchain direct queries,
+   and IPFS to provide complete agent/feedback information.
+
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import time
+import aiohttp
+from typing import Any, Dict, List, Optional, Union
+from datetime import datetime
+
+from .models import (
+    AgentId, ChainId, Address, URI, Timestamp,
+    AgentSummary, Feedback, SearchParams, SearchFeedbackParams
+)
+from .web3_client import Web3Client
+
+logger = logging.getLogger(__name__)
+
+
+class AgentIndexer:
+    """Indexer for agent discovery and search."""
+
+    def __init__(
+        self,
+        web3_client: Web3Client,
+        store: Optional[Any] = None,
+        embeddings: Optional[Any] = None,
+        subgraph_client: Optional[Any] = None,
+        identity_registry: Optional[Any] = None,
+        subgraph_url_overrides: Optional[Dict[int, str]] = None,
+    ):
+        """Initialize indexer with optional subgraph URL overrides for multiple chains."""
+        self.web3_client = web3_client
+        self.store = store or self._create_default_store()
+        self.embeddings = embeddings or self._create_default_embeddings()
+        self.subgraph_client = subgraph_client
+        self.identity_registry = identity_registry
+        self.subgraph_url_overrides = subgraph_url_overrides or {}
+        self._agent_cache = {}  # Cache for agent data
+        self._cache_timestamp = 0
+        self._cache_ttl = 7 * 24 * 60 * 60  # 1 week cache TTL (604800 seconds)
+        self._http_cache = {}  # Cache for HTTP content
+        self._http_cache_ttl = 60 * 60  # 1 hour cache TTL for HTTP content
+
+        # Cache for subgraph clients (one per chain)
+        self._subgraph_client_cache: Dict[int, Any] = {}
+
+        # If default subgraph_client provided, cache it for current chain
+        if self.subgraph_client:
+            self._subgraph_client_cache[self.web3_client.chain_id] = self.subgraph_client
+
+    def _create_default_store(self) -> Dict[str, Any]:
+        """Create default in-memory store."""
+        return {
+            "agents": {},
+            "feedback": {},
+            "embeddings": {},
+        }
+
+    def _create_default_embeddings(self):
+        """Create default embeddings model."""
+        try:
+            from sentence_transformers import SentenceTransformer
+            return SentenceTransformer('all-MiniLM-L6-v2')
+        except ImportError:
+            # Return None if sentence-transformers is not available
+            return None
+
+    async def _fetch_http_content(self, url: str) -> Optional[Dict[str, Any]]:
+        """Fetch content from HTTP/HTTPS URL with caching."""
+        # Check cache first
+        current_time = time.time()
+        if url in self._http_cache:
+            cached_data, timestamp = self._http_cache[url]
+            if current_time - timestamp < self._http_cache_ttl:
+                return cached_data
+        
+        try:
+            async with aiohttp.ClientSession() as session:
+                async with session.get(url, timeout=aiohttp.ClientTimeout(total=10)) as response:
+                    if response.status == 200:
+                        content = await response.json()
+                        # Cache the result
+                        self._http_cache[url] = (content, current_time)
+                        return content
+                    else:
+                        logger.warning(f"Failed to fetch {url}: HTTP {response.status}")
+                        return None
+        except Exception as e:
+            logger.warning(f"Error fetching HTTPS content from {url}: {e}")
+            return None
+
+    def _detect_uri_type(self, uri: str) -> str:
+        """Detect URI type (ipfs, https, http, unknown)."""
+        if uri.startswith("ipfs://"):
+            return "ipfs"
+        elif uri.startswith("https://"):
+            return "https"
+        elif uri.startswith("http://"):
+            return "http"
+        elif self._is_ipfs_cid(uri):
+            return "ipfs"
+        else:
+            return "unknown"
+
+    def _is_ipfs_cid(self, uri: str) -> bool:
+        """Check if string is an IPFS CID (without ipfs:// prefix)."""
+        # Basic IPFS CID patterns
+        # Qm... (CIDv0, 46 characters)
+        # bafy... (CIDv1, starts with bafy)
+        # bafk... (CIDv1, starts with bafk)
+        # bafg... (CIDv1, starts with bafg)
+        # bafh... (CIDv1, starts with bafh)
+        # bafq... (CIDv1, starts with bafq)
+        # bafr... (CIDv1, starts with bafr)
+        # bafs... (CIDv1, starts with bafs)
+        # baft... (CIDv1, starts with baft)
+        # bafu... (CIDv1, starts with bafu)
+        # bafv... (CIDv1, starts with bafv)
+        # bafw... (CIDv1, starts with bafw)
+        # bafx... (CIDv1, starts with bafx)
+        # bafy... (CIDv1, starts with bafy)
+        # bafz... (CIDv1, starts with bafz)
+        
+        if not uri:
+            return False
+            
+        # Check for CIDv0 (Qm...)
+        if uri.startswith("Qm") and len(uri) == 46:
+            return True
+            
+        # Check for CIDv1 (baf...)
+        # CIDv1 has variable length but typically 50+ characters
+        # We'll be more lenient for shorter CIDs that start with baf
+        if uri.startswith("baf") and len(uri) >= 8:
+            return True
+            
+        return False
+
+    def _is_ipfs_gateway_url(self, url: str) -> bool:
+        """Check if URL is an IPFS gateway URL."""
+        ipfs_gateways = [
+            "ipfs.io",
+            "gateway.pinata.cloud",
+            "cloudflare-ipfs.com",
+            "dweb.link",
+            "ipfs.fleek.co"
+        ]
+        return any(gateway in url for gateway in ipfs_gateways)
+
+    def _convert_gateway_to_ipfs(self, url: str) -> Optional[str]:
+        """Convert IPFS gateway URL to ipfs:// format."""
+        if "/ipfs/" in url:
+            # Extract hash from gateway URL
+            parts = url.split("/ipfs/")
+            if len(parts) == 2:
+                hash_part = parts[1].split("/")[0]  # Remove any path after hash
+                return f"ipfs://{hash_part}"
+        return None
+
+    async def _fetch_registration_file(self, uri: str) -> Optional[Dict[str, Any]]:
+        """Fetch registration file from IPFS or HTTPS."""
+        uri_type = self._detect_uri_type(uri)
+        
+        if uri_type == "ipfs":
+            # Normalize bare CID to ipfs:// format
+            if not uri.startswith("ipfs://"):
+                uri = f"ipfs://{uri}"
+            
+            # Use existing IPFS client (if available)
+            # For now, return None as IPFS fetching is handled by subgraph
+            return None
+        elif uri_type in ["https", "http"]:
+            # Check if it's an IPFS gateway URL
+            if self._is_ipfs_gateway_url(uri):
+                ipfs_uri = self._convert_gateway_to_ipfs(uri)
+                if ipfs_uri:
+                    # Try to fetch as IPFS first
+                    return await self._fetch_registration_file(ipfs_uri)
+            
+            # Fetch directly from HTTPS
+            return await self._fetch_http_content(uri)
+        else:
+            logger.warning(f"Unsupported URI type: {uri}")
+            return None
+
+    async def _fetch_feedback_file(self, uri: str) -> Optional[Dict[str, Any]]:
+        """Fetch feedback file from IPFS or HTTPS."""
+        uri_type = self._detect_uri_type(uri)
+        
+        if uri_type == "ipfs":
+            # Normalize bare CID to ipfs:// format
+            if not uri.startswith("ipfs://"):
+                uri = f"ipfs://{uri}"
+            
+            # Use existing IPFS client (if available)
+            # For now, return None as IPFS fetching is handled by subgraph
+            return None
+        elif uri_type in ["https", "http"]:
+            # Check if it's an IPFS gateway URL
+            if self._is_ipfs_gateway_url(uri):
+                ipfs_uri = self._convert_gateway_to_ipfs(uri)
+                if ipfs_uri:
+                    # Try to fetch as IPFS first
+                    return await self._fetch_feedback_file(ipfs_uri)
+            
+            # Fetch directly from HTTPS
+            return await self._fetch_http_content(uri)
+        else:
+            logger.warning(f"Unsupported URI type: {uri}")
+            return None
+
+    async def refresh_agent(self, agent_id: AgentId, deep: bool = False) -> AgentSummary:
+        """Refresh index for a single agent."""
+        # Parse agent ID
+        if ":" in agent_id:
+            chain_id, token_id = agent_id.split(":", 1)
+        else:
+            chain_id = self.web3_client.chain_id
+            token_id = agent_id
+
+        # Get basic agent data from contract
+        try:
+            if self.identity_registry:
+                token_uri = self.web3_client.call_contract(
+                    self.identity_registry,
+                    "tokenURI",
+                    int(token_id)
+                )
+            else:
+                raise ValueError("Identity registry not available")
+        except Exception as e:
+            raise ValueError(f"Failed to get agent data: {e}")
+
+        # Load registration file
+        registration_data = await self._load_registration_data(token_uri)
+        
+        # Create agent summary
+        summary = self._create_agent_summary(
+            chain_id=int(chain_id),
+            agent_id=agent_id,
+            registration_data=registration_data
+        )
+
+        # Store in index
+        self.store["agents"][agent_id] = summary
+
+        # Deep refresh if requested
+        if deep:
+            await self._deep_refresh_agent(summary)
+
+        return summary
+
+    async def refresh_agents(
+        self,
+        agent_ids: Optional[List[AgentId]] = None,
+        concurrency: int = 8,
+    ) -> List[AgentSummary]:
+        """Refresh index for multiple agents."""
+        if agent_ids is None:
+            # Get all known agents (this would need to be implemented)
+            agent_ids = list(self.store["agents"].keys())
+
+        # Use semaphore to limit concurrency
+        semaphore = asyncio.Semaphore(concurrency)
+
+        async def refresh_single(agent_id: AgentId) -> AgentSummary:
+            async with semaphore:
+                return await self.refresh_agent(agent_id)
+
+        # Execute all refreshes concurrently
+        tasks = [refresh_single(agent_id) for agent_id in agent_ids]
+        results = await asyncio.gather(*tasks, return_exceptions=True)
+
+        # Filter out exceptions
+        summaries = []
+        for result in results:
+            if isinstance(result, Exception):
+                logger.warning(f"Error refreshing agent: {result}")
+            else:
+                summaries.append(result)
+
+        return summaries
+
+    async def _load_registration_data(self, uri: str) -> Dict[str, Any]:
+        """Load registration data from URI."""
+        registration_file = await self._fetch_registration_file(uri)
+        if registration_file is None:
+            raise ValueError(f"Failed to load registration data from: {uri}")
+        return registration_file
+
+    def _create_agent_summary(
+        self,
+        chain_id: int,
+        agent_id: AgentId,
+        registration_data: Dict[str, Any]
+    ) -> AgentSummary:
+        """Create agent summary from registration data."""
+        # Extract endpoints
+        endpoints = registration_data.get("endpoints", [])
+        mcp = any(ep.get("name") == "MCP" for ep in endpoints)
+        a2a = any(ep.get("name") == "A2A" for ep in endpoints)
+        
+        ens = None
+        did = None
+        for ep in endpoints:
+            if ep.get("name") == "ENS":
+                ens = ep.get("endpoint")
+            elif ep.get("name") == "DID":
+                did = ep.get("endpoint")
+
+        # Extract capabilities (would need MCP/A2A crawling)
+        a2a_skills = []
+        mcp_tools = []
+        mcp_prompts = []
+        mcp_resources = []
+
+        return AgentSummary(
+            chainId=chain_id,
+            agentId=agent_id,
+            name=registration_data.get("name", ""),
+            image=registration_data.get("image"),
+            description=registration_data.get("description", ""),
+            owners=[],  # Would be populated from contract
+            operators=[],  # Would be populated from contract
+            mcp=mcp,
+            a2a=a2a,
+            ens=ens,
+            did=did,
+            walletAddress=registration_data.get("walletAddress"),
+            supportedTrusts=registration_data.get("supportedTrust", []),
+            a2aSkills=a2a_skills,
+            mcpTools=mcp_tools,
+            mcpPrompts=mcp_prompts,
+            mcpResources=mcp_resources,
+            active=registration_data.get("active", True),
+            extras={}
+        )
+
+    async def _deep_refresh_agent(self, summary: AgentSummary):
+        """Perform deep refresh of agent capabilities."""
+        # This would crawl MCP/A2A endpoints to extract capabilities
+        # For now, it's a placeholder
+        pass
+
+    def get_agent(self, agent_id: AgentId) -> AgentSummary:
+        """Get agent summary from index."""
+        # Parse chainId from agentId
+        chain_id, token_id = self._parse_agent_id(agent_id)
+        
+        # Get subgraph client for the chain
+        subgraph_client = None
+        full_agent_id = agent_id
+        
+        if chain_id is not None:
+            subgraph_client = self._get_subgraph_client_for_chain(chain_id)
+        else:
+            # No chainId in agentId, use SDK's default
+            # Construct full agentId format for subgraph query
+            default_chain_id = self.web3_client.chain_id
+            full_agent_id = f"{default_chain_id}:{token_id}"
+            subgraph_client = self.subgraph_client
+        
+        # Use subgraph if available (preferred)
+        if subgraph_client:
+            return self._get_agent_from_subgraph(full_agent_id, subgraph_client)
+        
+        # Fallback to local cache
+        if agent_id not in self.store["agents"]:
+            raise ValueError(f"Agent {agent_id} not found in index")
+        return self.store["agents"][agent_id]
+    
+    def _get_agent_from_subgraph(self, agent_id: AgentId, subgraph_client: Optional[Any] = None) -> AgentSummary:
+        """Get agent summary from subgraph."""
+        # Use provided client or default
+        client = subgraph_client or self.subgraph_client
+        if not client:
+            raise ValueError("No subgraph client available")
+        
+        try:
+            agent_data = client.get_agent_by_id(agent_id)
+            
+            if agent_data is None:
+                raise ValueError(f"Agent {agent_id} not found in subgraph")
+            
+            reg_file = agent_data.get('registrationFile') or {}
+            if not isinstance(reg_file, dict):
+                reg_file = {}
+            
+            return AgentSummary(
+                chainId=int(agent_data.get('chainId', 0)),
+                agentId=agent_data.get('id', agent_id),
+                name=reg_file.get('name', f"Agent {agent_id}"),
+                image=reg_file.get('image'),
+                description=reg_file.get('description', ''),
+                owners=[agent_data.get('owner', '')],
+                operators=agent_data.get('operators', []),
+                mcp=reg_file.get('mcpEndpoint') is not None,
+                a2a=reg_file.get('a2aEndpoint') is not None,
+                ens=reg_file.get('ens'),
+                did=reg_file.get('did'),
+                walletAddress=reg_file.get('agentWallet'),
+                supportedTrusts=reg_file.get('supportedTrusts', []),
+                a2aSkills=reg_file.get('a2aSkills', []),
+                mcpTools=reg_file.get('mcpTools', []),
+                mcpPrompts=reg_file.get('mcpPrompts', []),
+                mcpResources=reg_file.get('mcpResources', []),
+                active=reg_file.get('active', True),
+                x402support=reg_file.get('x402support', False),
+                extras={}
+            )
+            
+        except Exception as e:
+            raise ValueError(f"Failed to get agent from subgraph: {e}")
+
+    def search_agents(
+        self,
+        params: SearchParams,
+        sort: List[str],
+        page_size: int,
+        cursor: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Search for agents by querying the subgraph or blockchain."""
+        # Handle "all" chains shorthand
+        if params.chains == "all":
+            params.chains = self._get_all_configured_chains()
+            logger.info(f"Expanding 'all' to configured chains: {params.chains}")
+
+        # If chains are explicitly specified (even a single chain), use multi-chain path
+        # This ensures the correct subgraph client is used for the requested chain(s)
+        if params.chains and len(params.chains) > 0:
+            # Validate chains are configured
+            available_chains = set(self._get_all_configured_chains())
+            requested_chains = set(params.chains)
+            invalid_chains = requested_chains - available_chains
+
+            if invalid_chains:
+                logger.warning(
+                    f"Requested chains not configured: {invalid_chains}. "
+                    f"Available chains: {available_chains}"
+                )
+                # Filter to valid chains only
+                valid_chains = list(requested_chains & available_chains)
+                if not valid_chains:
+                    return {
+                        "items": [],
+                        "nextCursor": None,
+                        "meta": {
+                            "chains": list(requested_chains),
+                            "successfulChains": [],
+                            "failedChains": list(requested_chains),
+                            "error": f"No valid chains configured. Available: {list(available_chains)}"
+                        }
+                    }
+                params.chains = valid_chains
+
+            return asyncio.run(
+                self._search_agents_across_chains(params, sort, page_size, cursor)
+            )
+
+        # Use subgraph if available (preferred)
+        if self.subgraph_client:
+            return self._search_agents_via_subgraph(params, sort, page_size, cursor)
+
+        # Fallback to blockchain queries
+        return self._search_agents_via_blockchain(params, sort, page_size, cursor)
+
+    async def _search_agents_across_chains(
+        self,
+        params: SearchParams,
+        sort: List[str],
+        page_size: int,
+        cursor: Optional[str] = None,
+        timeout: float = 30.0,
+    ) -> Dict[str, Any]:
+        """
+        Search agents across multiple chains in parallel.
+
+        This method is called when params.chains contains 2+ chain IDs.
+        It executes one subgraph query per chain, all in parallel using asyncio.
+
+        Args:
+            params: Search parameters
+            sort: Sort specification
+            page_size: Number of results per page
+            cursor: Pagination cursor
+            timeout: Maximum time in seconds for all chain queries (default: 30.0)
+
+        Returns:
+            {
+                "items": [agent_dict, ...],
+                "nextCursor": str or None,
+                "meta": {
+                    "chains": [chainId, ...],
+                    "successfulChains": [chainId, ...],
+                    "failedChains": [chainId, ...],
+                    "totalResults": int,
+                    "timing": {"totalMs": int}
+                }
+            }
+        """
+        import time
+        start_time = time.time()
+        # Step 1: Determine which chains to query
+        chains_to_query = params.chains if params.chains else self._get_all_configured_chains()
+
+        if not chains_to_query or len(chains_to_query) == 0:
+            logger.warning("No chains specified or configured for multi-chain query")
+            return {"items": [], "nextCursor": None, "meta": {"chains": [], "successfulChains": [], "failedChains": []}}
+
+        # Step 2: Parse pagination cursor (if any)
+        chain_cursors = self._parse_multi_chain_cursor(cursor)
+        global_offset = chain_cursors.get("_global_offset", 0)
+
+        # Step 3: Define async function for querying a single chain
+        async def query_single_chain(chain_id: int) -> Dict[str, Any]:
+            """Query one chain and return its results with metadata."""
+            try:
+                # Get subgraph client for this chain
+                subgraph_client = self._get_subgraph_client_for_chain(chain_id)
+
+                if subgraph_client is None:
+                    logger.warning(f"No subgraph client available for chain {chain_id}")
+                    return {
+                        "chainId": chain_id,
+                        "status": "unavailable",
+                        "agents": [],
+                        "error": f"No subgraph configured for chain {chain_id}"
+                    }
+
+                # Build WHERE clause for this chain's query
+                # (reuse existing logic from _search_agents_via_subgraph)
+                where_clause = {}
+                reg_file_where = {}
+
+                if params.name is not None:
+                    reg_file_where["name_contains"] = params.name
+                if params.active is not None:
+                    reg_file_where["active"] = params.active
+                if params.x402support is not None:
+                    reg_file_where["x402support"] = params.x402support
+                if params.mcp is not None:
+                    if params.mcp:
+                        reg_file_where["mcpEndpoint_not"] = None
+                    else:
+                        reg_file_where["mcpEndpoint"] = None
+                if params.a2a is not None:
+                    if params.a2a:
+                        reg_file_where["a2aEndpoint_not"] = None
+                    else:
+                        reg_file_where["a2aEndpoint"] = None
+                if params.ens is not None:
+                    reg_file_where["ens"] = params.ens
+                if params.did is not None:
+                    reg_file_where["did"] = params.did
+                if params.walletAddress is not None:
+                    reg_file_where["agentWallet"] = params.walletAddress
+
+                if reg_file_where:
+                    where_clause["registrationFile_"] = reg_file_where
+
+                # Owner filtering
+                if params.owners is not None and len(params.owners) > 0:
+                    normalized_owners = [owner.lower() for owner in params.owners]
+                    if len(normalized_owners) == 1:
+                        where_clause["owner"] = normalized_owners[0]
+                    else:
+                        where_clause["owner_in"] = normalized_owners
+
+                # Operator filtering
+                if params.operators is not None and len(params.operators) > 0:
+                    normalized_operators = [op.lower() for op in params.operators]
+                    where_clause["operators_contains"] = normalized_operators
+
+                # Get pagination offset for this chain (not used in multi-chain, fetch all)
+                skip = 0
+
+                # Execute subgraph query
+                agents = subgraph_client.get_agents(
+                    where=where_clause if where_clause else None,
+                    first=page_size * 3,  # Fetch extra to allow for filtering/sorting
+                    skip=skip,
+                    order_by=self._extract_order_by(sort),
+                    order_direction=self._extract_order_direction(sort)
+                )
+
+                logger.info(f"Chain {chain_id}: fetched {len(agents)} agents")
+
+                return {
+                    "chainId": chain_id,
+                    "status": "success",
+                    "agents": agents,
+                    "count": len(agents),
+                }
+
+            except Exception as e:
+                logger.error(f"Error querying chain {chain_id}: {e}", exc_info=True)
+                return {
+                    "chainId": chain_id,
+                    "status": "error",
+                    "agents": [],
+                    "error": str(e)
+                }
+
+        # Step 4: Execute all chain queries in parallel with timeout
+        logger.info(f"Querying {len(chains_to_query)} chains in parallel: {chains_to_query}")
+        tasks = [query_single_chain(chain_id) for chain_id in chains_to_query]
+
+        try:
+            chain_results = await asyncio.wait_for(
+                asyncio.gather(*tasks),
+                timeout=timeout
+            )
+        except asyncio.TimeoutError:
+            logger.error(f"Multi-chain query timed out after {timeout}s")
+            # Collect results from completed tasks
+            chain_results = []
+            for task in tasks:
+                if task.done():
+                    try:
+                        chain_results.append(task.result())
+                    except Exception as e:
+                        logger.warning(f"Task failed: {e}")
+                else:
+                    # Task didn't complete - mark as timeout
+                    chain_results.append({
+                        "chainId": None,
+                        "status": "timeout",
+                        "agents": [],
+                        "error": f"Query timed out after {timeout}s"
+                    })
+
+        # Step 5: Extract successful results and track failures
+        all_agents = []
+        successful_chains = []
+        failed_chains = []
+
+        for result in chain_results:
+            chain_id = result["chainId"]
+
+            if result["status"] == "success":
+                successful_chains.append(chain_id)
+                all_agents.extend(result["agents"])
+            else:
+                failed_chains.append(chain_id)
+                logger.warning(
+                    f"Chain {chain_id} query failed: {result.get('error', 'Unknown error')}"
+                )
+
+        logger.info(f"Multi-chain query: {len(successful_chains)} successful, {len(failed_chains)} failed, {len(all_agents)} total agents")
+
+        # If ALL chains failed, raise error
+        if len(successful_chains) == 0:
+            raise ConnectionError(
+                f"All chains failed: {', '.join(str(c) for c in failed_chains)}"
+            )
+
+        # Step 6: Apply cross-chain filtering (for fields not supported by subgraph WHERE clause)
+        filtered_agents = self._apply_cross_chain_filters(all_agents, params)
+        logger.info(f"After cross-chain filters: {len(filtered_agents)} agents")
+
+        # Step 7: Deduplicate if requested
+        deduplicated_agents = self._deduplicate_agents_cross_chain(filtered_agents, params)
+        logger.info(f"After deduplication: {len(deduplicated_agents)} agents")
+
+        # Step 8: Sort across chains
+        sorted_agents = self._sort_agents_cross_chain(deduplicated_agents, sort)
+        logger.info(f"After sorting: {len(sorted_agents)} agents")
+
+        # Step 9: Apply pagination
+        start_idx = global_offset
+        paginated_agents = sorted_agents[start_idx:start_idx + page_size]
+
+        # Step 10: Convert to result format (keep as dicts, SDK will convert to AgentSummary)
+        results = []
+        for agent_data in paginated_agents:
+            reg_file = agent_data.get('registrationFile') or {}
+            if not isinstance(reg_file, dict):
+                reg_file = {}
+
+            result_agent = {
+                "agentId": agent_data.get('id'),
+                "chainId": agent_data.get('chainId'),
+                "name": reg_file.get('name', f"Agent {agent_data.get('agentId')}"),
+                "description": reg_file.get('description', ''),
+                "image": reg_file.get('image'),
+                "owner": agent_data.get('owner'),
+                "operators": agent_data.get('operators', []),
+                "mcp": reg_file.get('mcpEndpoint') is not None,
+                "a2a": reg_file.get('a2aEndpoint') is not None,
+                "ens": reg_file.get('ens'),
+                "did": reg_file.get('did'),
+                "walletAddress": reg_file.get('agentWallet'),
+                "supportedTrusts": reg_file.get('supportedTrusts', []),
+                "a2aSkills": reg_file.get('a2aSkills', []),
+                "mcpTools": reg_file.get('mcpTools', []),
+                "mcpPrompts": reg_file.get('mcpPrompts', []),
+                "mcpResources": reg_file.get('mcpResources', []),
+                "active": reg_file.get('active', True),
+                "x402support": reg_file.get('x402support', False),
+                "totalFeedback": agent_data.get('totalFeedback', 0),
+                "lastActivity": agent_data.get('lastActivity'),
+                "updatedAt": agent_data.get('updatedAt'),
+                "extras": {}
+            }
+
+            # Add deployedOn if deduplication was used
+            if 'deployedOn' in agent_data:
+                result_agent['extras']['deployedOn'] = agent_data['deployedOn']
+
+            results.append(result_agent)
+
+        # Step 11: Calculate next cursor
+        next_cursor = None
+        if len(sorted_agents) > start_idx + page_size:
+            # More results available
+            next_cursor = self._create_multi_chain_cursor(
+                global_offset=start_idx + page_size
+            )
+
+        # Step 12: Build response with metadata
+        query_time = time.time() - start_time
+
+        return {
+            "items": results,
+            "nextCursor": next_cursor,
+            "meta": {
+                "chains": chains_to_query,
+                "successfulChains": successful_chains,
+                "failedChains": failed_chains,
+                "totalResults": len(sorted_agents),
+                "pageResults": len(results),
+                "timing": {
+                    "totalMs": int(query_time * 1000),
+                    "averagePerChainMs": int(query_time * 1000 / len(chains_to_query)) if chains_to_query else 0,
+                }
+            }
+        }
+
+    def _search_agents_via_subgraph(
+        self,
+        params: SearchParams,
+        sort: List[str],
+        page_size: int,
+        cursor: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Search for agents using the subgraph."""
+        # Build subgraph query filters
+        where_clause = {}
+        reg_file_where = {}
+        
+        if params.name is not None:
+            reg_file_where["name_contains"] = params.name
+        if params.active is not None:
+            reg_file_where["active"] = params.active
+        if params.x402support is not None:
+            reg_file_where["x402support"] = params.x402support
+        if params.mcp is not None:
+            if params.mcp:
+                reg_file_where["mcpEndpoint_not"] = None
+            else:
+                reg_file_where["mcpEndpoint"] = None
+        if params.a2a is not None:
+            if params.a2a:
+                reg_file_where["a2aEndpoint_not"] = None
+            else:
+                reg_file_where["a2aEndpoint"] = None
+        if params.ens is not None:
+            reg_file_where["ens"] = params.ens
+        if params.did is not None:
+            reg_file_where["did"] = params.did
+        if params.walletAddress is not None:
+            reg_file_where["agentWallet"] = params.walletAddress
+
+        if reg_file_where:
+            where_clause["registrationFile_"] = reg_file_where
+
+        # Owner filtering
+        if params.owners is not None and len(params.owners) > 0:
+            # Normalize addresses to lowercase for case-insensitive matching
+            normalized_owners = [owner.lower() for owner in params.owners]
+            if len(normalized_owners) == 1:
+                where_clause["owner"] = normalized_owners[0]
+            else:
+                where_clause["owner_in"] = normalized_owners
+
+        # Operator filtering 
+        if params.operators is not None and len(params.operators) > 0:
+            # Normalize addresses to lowercase for case-insensitive matching
+            normalized_operators = [op.lower() for op in params.operators]
+            # For operators (array field), use contains to check if any operator matches
+            where_clause["operators_contains"] = normalized_operators
+        
+        # Calculate pagination
+        skip = 0
+        if cursor:
+            try:
+                skip = int(cursor)
+            except ValueError:
+                skip = 0
+        
+        # Determine sort
+        order_by = "createdAt"
+        order_direction = "desc"
+        if sort and len(sort) > 0:
+            sort_field = sort[0].split(":")
+            if len(sort_field) >= 1:
+                order_by = sort_field[0]
+            if len(sort_field) >= 2:
+                order_direction = sort_field[1]
+        
+        try:
+            agents = self.subgraph_client.get_agents(
+                where=where_clause if where_clause else None,
+                first=page_size,
+                skip=skip,
+                order_by=order_by,
+                order_direction=order_direction
+            )
+            
+            results = []
+            for agent in agents:
+                reg_file = agent.get('registrationFile') or {}
+                # Ensure reg_file is a dict
+                if not isinstance(reg_file, dict):
+                    reg_file = {}
+                    
+                agent_data = {
+                    "agentId": agent.get('id'),
+                    "chainId": agent.get('chainId'),
+                    "name": reg_file.get('name', f"Agent {agent.get('agentId')}"),
+                    "description": reg_file.get('description', ''),
+                    "image": reg_file.get('image'),
+                    "owner": agent.get('owner'),
+                    "operators": agent.get('operators', []),
+                    "mcp": reg_file.get('mcpEndpoint') is not None,
+                    "a2a": reg_file.get('a2aEndpoint') is not None,
+                    "ens": reg_file.get('ens'),
+                    "did": reg_file.get('did'),
+                    "walletAddress": reg_file.get('agentWallet'),
+                    "supportedTrusts": reg_file.get('supportedTrusts', []),
+                    "a2aSkills": reg_file.get('a2aSkills', []),
+                    "mcpTools": reg_file.get('mcpTools', []),
+                    "mcpPrompts": reg_file.get('mcpPrompts', []),
+                    "mcpResources": reg_file.get('mcpResources', []),
+                    "active": reg_file.get('active', True),
+                    "x402support": reg_file.get('x402support', False),
+                    "totalFeedback": agent.get('totalFeedback', 0),
+                    "lastActivity": agent.get('lastActivity'),
+                    "updatedAt": agent.get('updatedAt'),
+                    "extras": {}
+                }
+                
+                if params.chains is not None:
+                    if agent_data["chainId"] not in params.chains:
+                        continue
+                if params.supportedTrust is not None:
+                    if not any(trust in agent_data["supportedTrusts"] for trust in params.supportedTrust):
+                        continue
+                
+                results.append(agent_data)
+            
+            next_cursor = str(skip + len(results)) if len(results) == page_size else None
+            return {"items": results, "nextCursor": next_cursor}
+            
+        except Exception as e:
+            logger.warning(f"Subgraph search failed: {e}")
+            return {"items": [], "nextCursor": None}
+    
+    def _search_agents_via_blockchain(
+        self,
+        params: SearchParams,
+        sort: List[str],
+        page_size: int,
+        cursor: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Search for agents by querying the blockchain (fallback)."""
+        return {"items": [], "nextCursor": None}
+
+    def _apply_filters(self, agents: List[Dict[str, Any]], params: SearchParams) -> List[Dict[str, Any]]:
+        """Apply search filters to agents."""
+        filtered = agents
+        
+        if params.chains is not None:
+            filtered = [a for a in filtered if a.get("chainId") in params.chains]
+        
+        if params.name is not None:
+            filtered = [a for a in filtered if params.name.lower() in a.get("name", "").lower()]
+        
+        if params.description is not None:
+            # This would use semantic search with embeddings
+            filtered = [a for a in filtered if params.description.lower() in a.get("description", "").lower()]
+        
+        if params.owners is not None:
+            filtered = [a for a in filtered if any(owner in params.owners for owner in a.get("owners", []))]
+        
+        if params.operators is not None:
+            filtered = [a for a in filtered if any(op in params.operators for op in a.get("operators", []))]
+        
+        if params.mcp is not None:
+            filtered = [a for a in filtered if a.get("mcp") == params.mcp]
+        
+        if params.a2a is not None:
+            filtered = [a for a in filtered if a.get("a2a") == params.a2a]
+        
+        if params.ens is not None:
+            filtered = [a for a in filtered if a.get("ens") and params.ens.lower() in a.get("ens", "").lower()]
+        
+        if params.did is not None:
+            filtered = [a for a in filtered if a.get("did") == params.did]
+        
+        if params.walletAddress is not None:
+            filtered = [a for a in filtered if a.get("walletAddress") == params.walletAddress]
+        
+        if params.supportedTrust is not None:
+            filtered = [a for a in filtered if any(trust in params.supportedTrust for trust in a.get("supportedTrusts", []))]
+        
+        if params.a2aSkills is not None:
+            filtered = [a for a in filtered if any(skill in params.a2aSkills for skill in a.get("a2aSkills", []))]
+        
+        if params.mcpTools is not None:
+            filtered = [a for a in filtered if any(tool in params.mcpTools for tool in a.get("mcpTools", []))]
+        
+        if params.mcpPrompts is not None:
+            filtered = [a for a in filtered if any(prompt in params.mcpPrompts for prompt in a.get("mcpPrompts", []))]
+        
+        if params.mcpResources is not None:
+            filtered = [a for a in filtered if any(resource in params.mcpResources for resource in a.get("mcpResources", []))]
+        
+        if params.active is not None:
+            filtered = [a for a in filtered if a.get("active") == params.active]
+        
+        if params.x402support is not None:
+            filtered = [a for a in filtered if a.get("x402support") == params.x402support]
+        
+        return filtered
+
+    def _apply_sorting(self, agents: List[AgentSummary], sort: List[str]) -> List[AgentSummary]:
+        """Apply sorting to agents."""
+        def sort_key(agent):
+            key_values = []
+            for sort_field in sort:
+                field, direction = sort_field.split(":", 1)
+                if hasattr(agent, field):
+                    value = getattr(agent, field)
+                    if direction == "desc":
+                        value = -value if isinstance(value, (int, float)) else value
+                    key_values.append(value)
+            return key_values
+        
+        return sorted(agents, key=sort_key)
+
+    def get_feedback(
+        self,
+        agentId: AgentId,
+        clientAddress: Address,
+        feedbackIndex: int,
+    ) -> Feedback:
+        """Get single feedback by agent ID, client address, and index."""
+        # Use subgraph if available (preferred)
+        if self.subgraph_client:
+            return self._get_feedback_from_subgraph(agentId, clientAddress, feedbackIndex)
+        
+        # Fallback to local store (if populated in future)
+        # For now, raise error if subgraph unavailable
+        feedback_id = Feedback.create_id(agentId, clientAddress, feedbackIndex)
+        if feedback_id not in self.store["feedback"]:
+            raise ValueError(f"Feedback {feedback_id} not found (subgraph required)")
+        return self.store["feedback"][feedback_id]
+    
+    def _get_feedback_from_subgraph(
+        self,
+        agentId: AgentId,
+        clientAddress: Address,
+        feedbackIndex: int,
+    ) -> Feedback:
+        """Get feedback from subgraph."""
+        # Normalize addresses to lowercase for consistent storage
+        normalized_client_address = self.web3_client.normalize_address(clientAddress)
+        
+        # Build feedback ID in format: chainId:agentId:clientAddress:feedbackIndex
+        if ":" in agentId:
+            feedback_id = f"{agentId}:{normalized_client_address}:{feedbackIndex}"
+        else:
+            chain_id = str(self.web3_client.chain_id)
+            feedback_id = f"{chain_id}:{agentId}:{normalized_client_address}:{feedbackIndex}"
+        
+        try:
+            feedback_data = self.subgraph_client.get_feedback_by_id(feedback_id)
+            
+            if feedback_data is None:
+                raise ValueError(f"Feedback {feedback_id} not found in subgraph")
+            
+            return self._map_subgraph_feedback_to_model(feedback_data, agentId, clientAddress, feedbackIndex)
+            
+        except Exception as e:
+            raise ValueError(f"Failed to get feedback from subgraph: {e}")
+    
+    def _map_subgraph_feedback_to_model(
+        self,
+        feedback_data: Dict[str, Any],
+        agentId: AgentId,
+        clientAddress: Address,
+        feedbackIndex: int,
+    ) -> Feedback:
+        """Map subgraph feedback data to Feedback model."""
+        feedback_file = feedback_data.get('feedbackFile') or {}
+        if not isinstance(feedback_file, dict):
+            feedback_file = {}
+        
+        # Map responses
+        responses_data = feedback_data.get('responses', [])
+        answers = []
+        for resp in responses_data:
+            answers.append({
+                'responder': resp.get('responder'),
+                'responseUri': resp.get('responseUri'),
+                'responseHash': resp.get('responseHash'),
+                'createdAt': resp.get('createdAt')
+            })
+        
+        # Map tags - check if they're hex bytes32 or plain strings
+        tags = []
+        tag1 = feedback_data.get('tag1') or feedback_file.get('tag1')
+        tag2 = feedback_data.get('tag2') or feedback_file.get('tag2')
+        
+        # Convert hex bytes32 to readable tags
+        if tag1 or tag2:
+            tags = self._hexBytes32ToTags(
+                tag1 if isinstance(tag1, str) else "",
+                tag2 if isinstance(tag2, str) else ""
+            )
+        
+        # If conversion failed, try as plain strings
+        if not tags:
+            if tag1 and not tag1.startswith("0x"):
+                tags.append(tag1)
+            if tag2 and not tag2.startswith("0x"):
+                tags.append(tag2)
+        
+        return Feedback(
+            id=Feedback.create_id(agentId, clientAddress, feedbackIndex),
+            agentId=agentId,
+            reviewer=self.web3_client.normalize_address(clientAddress),
+            score=feedback_data.get('score'),
+            tags=tags,
+            text=feedback_file.get('text'),
+            capability=feedback_file.get('capability'),
+            context=feedback_file.get('context'),
+            proofOfPayment={
+                'fromAddress': feedback_file.get('proofOfPaymentFromAddress'),
+                'toAddress': feedback_file.get('proofOfPaymentToAddress'),
+                'chainId': feedback_file.get('proofOfPaymentChainId'),
+                'txHash': feedback_file.get('proofOfPaymentTxHash'),
+            } if feedback_file.get('proofOfPaymentFromAddress') else None,
+            fileURI=feedback_data.get('feedbackUri'),
+            createdAt=feedback_data.get('createdAt', int(time.time())),
+            answers=answers,
+            isRevoked=feedback_data.get('isRevoked', False),
+            name=feedback_file.get('name'),
+            skill=feedback_file.get('skill'),
+            task=feedback_file.get('task'),
+        )
+    
+    def search_feedback(
+        self,
+        agentId: AgentId,
+        clientAddresses: Optional[List[Address]] = None,
+        tags: Optional[List[str]] = None,
+        capabilities: Optional[List[str]] = None,
+        skills: Optional[List[str]] = None,
+        tasks: Optional[List[str]] = None,
+        names: Optional[List[str]] = None,
+        minScore: Optional[int] = None,
+        maxScore: Optional[int] = None,
+        include_revoked: bool = False,
+        first: int = 100,
+        skip: int = 0,
+    ) -> List[Feedback]:
+        """Search feedback for an agent - uses subgraph if available."""
+        # Parse chainId from agentId
+        chain_id, token_id = self._parse_agent_id(agentId)
+        
+        # Get subgraph client for the chain
+        subgraph_client = None
+        full_agent_id = agentId
+        
+        if chain_id is not None:
+            subgraph_client = self._get_subgraph_client_for_chain(chain_id)
+        else:
+            # No chainId in agentId, use SDK's default
+            # Construct full agentId format for subgraph query
+            default_chain_id = self.web3_client.chain_id
+            full_agent_id = f"{default_chain_id}:{token_id}"
+            subgraph_client = self.subgraph_client
+        
+        # Use subgraph if available (preferred)
+        if subgraph_client:
+            return self._search_feedback_subgraph(
+                full_agent_id, clientAddresses, tags, capabilities, skills, tasks, names,
+                minScore, maxScore, include_revoked, first, skip, subgraph_client
+            )
+        
+        # Fallback not implemented (would require blockchain queries)
+        # For now, return empty if subgraph unavailable
+        return []
+    
+    def _search_feedback_subgraph(
+        self,
+        agentId: AgentId,
+        clientAddresses: Optional[List[Address]],
+        tags: Optional[List[str]],
+        capabilities: Optional[List[str]],
+        skills: Optional[List[str]],
+        tasks: Optional[List[str]],
+        names: Optional[List[str]],
+        minScore: Optional[int],
+        maxScore: Optional[int],
+        include_revoked: bool,
+        first: int,
+        skip: int,
+        subgraph_client: Optional[Any] = None,
+    ) -> List[Feedback]:
+        """Search feedback using subgraph."""
+        # Use provided client or default
+        client = subgraph_client or self.subgraph_client
+        if not client:
+            return []
+        
+        # Create SearchFeedbackParams
+        params = SearchFeedbackParams(
+            agents=[agentId],
+            reviewers=clientAddresses,
+            tags=tags,
+            capabilities=capabilities,
+            skills=skills,
+            tasks=tasks,
+            names=names,
+            minScore=minScore,
+            maxScore=maxScore,
+            includeRevoked=include_revoked
+        )
+        
+        # Query subgraph
+        feedbacks_data = client.search_feedback(
+            params=params,
+            first=first,
+            skip=skip,
+            order_by="createdAt",
+            order_direction="desc"
+        )
+        
+        # Map to Feedback objects
+        feedbacks = []
+        for fb_data in feedbacks_data:
+            # Parse agentId from feedback ID
+            feedback_id = fb_data['id']
+            parts = feedback_id.split(':')
+            if len(parts) >= 2:
+                agent_id_str = f"{parts[0]}:{parts[1]}"
+                client_addr = parts[2] if len(parts) > 2 else ""
+                feedback_idx = int(parts[3]) if len(parts) > 3 else 1
+            else:
+                agent_id_str = feedback_id
+                client_addr = ""
+                feedback_idx = 1
+            
+            feedback = self._map_subgraph_feedback_to_model(
+                fb_data, agent_id_str, client_addr, feedback_idx
+            )
+            feedbacks.append(feedback)
+        
+        return feedbacks
+    
+    def _hexBytes32ToTags(self, tag1: str, tag2: str) -> List[str]:
+        """Convert hex bytes32 tags back to strings, or return plain strings as-is.
+        
+        The subgraph now stores tags as human-readable strings (not hex),
+        so this method handles both formats for backwards compatibility.
+        """
+        tags = []
+        
+        if tag1 and tag1 != "0x" + "00" * 32:
+            # If it's already a plain string (from subgraph), use it directly
+            if not tag1.startswith("0x"):
+                if tag1:
+                    tags.append(tag1)
+            else:
+                # Try to convert from hex bytes32 (on-chain format)
+                try:
+                    hex_bytes = bytes.fromhex(tag1[2:])
+                    tag1_str = hex_bytes.rstrip(b'\x00').decode('utf-8', errors='ignore')
+                    if tag1_str:
+                        tags.append(tag1_str)
+                except Exception:
+                    pass  # Ignore invalid hex strings
+        
+        if tag2 and tag2 != "0x" + "00" * 32:
+            # If it's already a plain string (from subgraph), use it directly
+            if not tag2.startswith("0x"):
+                if tag2:
+                    tags.append(tag2)
+            else:
+                # Try to convert from hex bytes32 (on-chain format)
+                try:
+                    if tag2.startswith("0x"):
+                        hex_bytes = bytes.fromhex(tag2[2:])
+                    else:
+                        hex_bytes = bytes.fromhex(tag2)
+                    tag2_str = hex_bytes.rstrip(b'\x00').decode('utf-8', errors='ignore')
+                    if tag2_str:
+                        tags.append(tag2_str)
+                except Exception:
+                    pass  # Ignore invalid hex strings
+        
+        return tags
+
+    def get_reputation_summary(
+        self,
+        agent_id: AgentId,
+        group_by: List[str],
+        reviewers: Optional[List[Address]] = None,
+        since: Optional[Timestamp] = None,
+        until: Optional[Timestamp] = None,
+        sort: List[str] = None,
+        page_size: int = 100,
+        cursor: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Get reputation summary for an agent."""
+        # This would aggregate feedback data
+        # For now, return empty result
+        return {
+            "groups": [],
+            "nextCursor": None
+        }
+
+    def get_reputation_map(
+        self,
+        agents: List[Union[AgentSummary, AgentId]],
+        filters: Dict[str, Any],
+        sort: List[str],
+        reviewers: Optional[List[Address]] = None,
+    ) -> List[Dict[str, Any]]:
+        """Get reputation map for multiple agents."""
+        # This would calculate reputation metrics for each agent
+        # For now, return empty result
+        return []
+
+    def _get_agent_from_blockchain(self, token_id: int, sdk) -> Optional[Dict[str, Any]]:
+        """Get agent data from blockchain."""
+        try:
+            # Get token URI from contract
+            token_uri = self.web3_client.call_contract(
+                sdk.identity_registry,
+                "tokenURI",
+                token_id
+            )
+            
+            # Get owner
+            owner = self.web3_client.call_contract(
+                sdk.identity_registry,
+                "ownerOf",
+                token_id
+            )
+            
+            # Create agent ID
+            agent_id = f"{sdk.chain_id}:{token_id}"
+            
+            # Try to load registration data from IPFS
+            registration_data = self._load_registration_from_ipfs(token_uri, sdk)
+            
+            if registration_data:
+                # Use data from IPFS
+                return {
+                    "agentId": agent_id,
+                    "name": registration_data.get("name", f"Agent {token_id}"),
+                    "description": registration_data.get("description", f"Agent registered with token ID {token_id}"),
+                    "owner": owner,
+                    "tokenId": token_id,
+                    "tokenURI": token_uri,
+                    "x402support": registration_data.get("x402support", False),
+                    "trustModels": registration_data.get("trustModels", ["reputation"]),
+                    "active": registration_data.get("active", True),
+                    "endpoints": registration_data.get("endpoints", []),
+                    "image": registration_data.get("image"),
+                    "walletAddress": registration_data.get("walletAddress"),
+                    "metadata": registration_data.get("metadata", {})
+                }
+            else:
+                # Fallback to basic data
+                return {
+                    "agentId": agent_id,
+                    "name": f"Agent {token_id}",
+                    "description": f"Agent registered with token ID {token_id}",
+                    "owner": owner,
+                    "tokenId": token_id,
+                    "tokenURI": token_uri,
+                    "x402support": False,
+                    "trustModels": ["reputation"],
+                    "active": True,
+                    "endpoints": [],
+                    "image": None,
+                    "walletAddress": None,
+                    "metadata": {}
+                }
+        except Exception as e:
+            logger.error(f"Error loading agent {token_id}: {e}")
+            return None
+
+    def _load_registration_from_ipfs(self, token_uri: str, sdk) -> Optional[Dict[str, Any]]:
+        """Load agent registration data from IPFS or HTTP gateway."""
+        try:
+            import json
+            import requests
+            
+            # Extract IPFS hash from token URI
+            if token_uri.startswith("ipfs://"):
+                ipfs_hash = token_uri[7:]  # Remove "ipfs://" prefix
+            elif token_uri.startswith("https://") and "ipfs" in token_uri:
+                # Extract hash from IPFS gateway URL
+                parts = token_uri.split("/")
+                ipfs_hash = parts[-1] if parts[-1] else parts[-2]
+            elif token_uri.startswith("https://"):
+                # Direct HTTP URL - try to fetch directly
+                try:
+                    response = requests.get(token_uri, timeout=10)
+                    response.raise_for_status()
+                    return response.json()
+                except Exception as e:
+                    logger.warning(f"Could not load HTTP data from {token_uri}: {e}")
+                    return None
+            else:
+                return None
+            
+            # Try local IPFS client first (if available)
+            if hasattr(sdk, 'ipfs_client') and sdk.ipfs_client is not None:
+                try:
+                    data = sdk.ipfs_client.get(ipfs_hash)
+                    if data:
+                        return json.loads(data)
+                except Exception as e:
+                    logger.warning(f"Could not load from local IPFS for {ipfs_hash}: {e}")
+            
+            # Fallback to IPFS HTTP gateways
+            gateways = [
+                f"https://ipfs.io/ipfs/{ipfs_hash}",
+                f"https://gateway.pinata.cloud/ipfs/{ipfs_hash}",
+                f"https://cloudflare-ipfs.com/ipfs/{ipfs_hash}",
+                f"https://dweb.link/ipfs/{ipfs_hash}"
+            ]
+            
+            for gateway_url in gateways:
+                try:
+                    response = requests.get(gateway_url, timeout=10)
+                    response.raise_for_status()
+                    return response.json()
+                except Exception as e:
+                    logger.debug(f"Could not load from {gateway_url}: {e}")
+                    continue
+            
+            logger.warning(f"Could not load data for {ipfs_hash} from any source")
+            return None
+                
+        except Exception as e:
+            logger.warning(f"Could not parse token URI {token_uri}: {e}")
+            return None
+
+    def _get_subgraph_client_for_chain(self, chain_id: int):
+        """
+        Get or create SubgraphClient for a specific chain.
+
+        Checks (in order):
+        1. Client cache (already created)
+        2. Subgraph URL overrides (from constructor)
+        3. DEFAULT_SUBGRAPH_URLS (from contracts.py)
+        4. Environment variables (SUBGRAPH_URL_<chainId>)
+
+        Returns None if no subgraph URL is available for this chain.
+        """
+        # Check cache first
+        if chain_id in self._subgraph_client_cache:
+            return self._subgraph_client_cache[chain_id]
+
+        # Get subgraph URL for this chain
+        subgraph_url = self._get_subgraph_url_for_chain(chain_id)
+
+        if subgraph_url is None:
+            logger.warning(f"No subgraph URL configured for chain {chain_id}")
+            return None
+
+        # Create new SubgraphClient
+        from .subgraph_client import SubgraphClient
+        client = SubgraphClient(subgraph_url)
+
+        # Cache for future use
+        self._subgraph_client_cache[chain_id] = client
+
+        logger.info(f"Created subgraph client for chain {chain_id}: {subgraph_url}")
+
+        return client
+
+    def _get_subgraph_url_for_chain(self, chain_id: int) -> Optional[str]:
+        """
+        Get subgraph URL for a specific chain.
+
+        Priority order:
+        1. Constructor-provided overrides (self.subgraph_url_overrides)
+        2. DEFAULT_SUBGRAPH_URLS from contracts.py
+        3. Environment variable SUBGRAPH_URL_<chainId>
+        4. None (not configured)
+        """
+        import os
+
+        # 1. Check constructor overrides
+        if chain_id in self.subgraph_url_overrides:
+            return self.subgraph_url_overrides[chain_id]
+
+        # 2. Check DEFAULT_SUBGRAPH_URLS
+        from .contracts import DEFAULT_SUBGRAPH_URLS
+        if chain_id in DEFAULT_SUBGRAPH_URLS:
+            return DEFAULT_SUBGRAPH_URLS[chain_id]
+
+        # 3. Check environment variable
+        env_key = f"SUBGRAPH_URL_{chain_id}"
+        env_url = os.environ.get(env_key)
+        if env_url:
+            logger.info(f"Using subgraph URL from environment: {env_key}={env_url}")
+            return env_url
+
+        # 4. Not found
+        return None
+
+    def _parse_agent_id(self, agent_id: AgentId) -> tuple[Optional[int], str]:
+        """
+        Parse agentId to extract chainId and tokenId.
+        
+        Returns:
+            (chain_id, token_id_str) where:
+            - chain_id: int if "chainId:tokenId" format, None if just "tokenId"
+            - token_id_str: the tokenId part (always present)
+        """
+        if ":" in agent_id:
+            parts = agent_id.split(":", 1)
+            try:
+                chain_id = int(parts[0])
+                token_id = parts[1]
+                return (chain_id, token_id)
+            except ValueError:
+                # Invalid chainId, treat as tokenId only
+                return (None, agent_id)
+        return (None, agent_id)
+
+    def _get_all_configured_chains(self) -> List[int]:
+        """
+        Get list of all chains that have subgraphs configured.
+
+        This is used when params.chains is None (query all available chains).
+        """
+        import os
+        from .contracts import DEFAULT_SUBGRAPH_URLS
+
+        chains = set()
+
+        # Add chains from DEFAULT_SUBGRAPH_URLS
+        chains.update(DEFAULT_SUBGRAPH_URLS.keys())
+
+        # Add chains from constructor overrides
+        chains.update(self.subgraph_url_overrides.keys())
+
+        # Add chains from environment variables
+        for key, value in os.environ.items():
+            if key.startswith("SUBGRAPH_URL_") and value:
+                try:
+                    chain_id = int(key.replace("SUBGRAPH_URL_", ""))
+                    chains.add(chain_id)
+                except ValueError:
+                    pass
+
+        return sorted(list(chains))
+
+    def _apply_cross_chain_filters(
+        self,
+        agents: List[Dict[str, Any]],
+        params: SearchParams
+    ) -> List[Dict[str, Any]]:
+        """
+        Apply filters that couldn't be expressed in subgraph WHERE clause.
+
+        Most filters are already applied by the subgraph query, but some
+        (like supportedTrust, mcpTools, etc.) need post-processing.
+        """
+        filtered = agents
+
+        # Filter by supportedTrust (if specified)
+        if params.supportedTrust is not None:
+            filtered = [
+                agent for agent in filtered
+                if any(
+                    trust in agent.get('registrationFile', {}).get('supportedTrusts', [])
+                    for trust in params.supportedTrust
+                )
+            ]
+
+        # Filter by mcpTools (if specified)
+        if params.mcpTools is not None:
+            filtered = [
+                agent for agent in filtered
+                if any(
+                    tool in agent.get('registrationFile', {}).get('mcpTools', [])
+                    for tool in params.mcpTools
+                )
+            ]
+
+        # Filter by a2aSkills (if specified)
+        if params.a2aSkills is not None:
+            filtered = [
+                agent for agent in filtered
+                if any(
+                    skill in agent.get('registrationFile', {}).get('a2aSkills', [])
+                    for skill in params.a2aSkills
+                )
+            ]
+
+        # Filter by mcpPrompts (if specified)
+        if params.mcpPrompts is not None:
+            filtered = [
+                agent for agent in filtered
+                if any(
+                    prompt in agent.get('registrationFile', {}).get('mcpPrompts', [])
+                    for prompt in params.mcpPrompts
+                )
+            ]
+
+        # Filter by mcpResources (if specified)
+        if params.mcpResources is not None:
+            filtered = [
+                agent for agent in filtered
+                if any(
+                    resource in agent.get('registrationFile', {}).get('mcpResources', [])
+                    for resource in params.mcpResources
+                )
+            ]
+
+        return filtered
+
+    def _deduplicate_agents_cross_chain(
+        self,
+        agents: List[Dict[str, Any]],
+        params: SearchParams
+    ) -> List[Dict[str, Any]]:
+        """
+        Deduplicate agents across chains (if requested).
+
+        Strategy:
+        - By default, DON'T deduplicate (agents on different chains are different entities)
+        - If params.deduplicate_cross_chain=True, deduplicate by (owner, registration_hash)
+
+        When deduplicating:
+        - Keep the first instance encountered
+        - Add 'deployedOn' array with all chain IDs where this agent exists
+        """
+        # Check if deduplication requested
+        if not params.deduplicate_cross_chain:
+            return agents
+
+        # Group agents by identity key
+        seen = {}
+        deduplicated = []
+
+        for agent in agents:
+            # Create identity key: (owner, name, description)
+            # This identifies "the same agent" across chains
+            owner = agent.get('owner', '').lower()
+            reg_file = agent.get('registrationFile', {})
+            name = reg_file.get('name', '')
+            description = reg_file.get('description', '')
+
+            identity_key = (owner, name, description)
+
+            if identity_key not in seen:
+                # First time seeing this agent
+                seen[identity_key] = agent
+
+                # Add deployedOn array
+                agent['deployedOn'] = [agent['chainId']]
+
+                deduplicated.append(agent)
+            else:
+                # Already seen this agent on another chain
+                # Add this chain to deployedOn array
+                seen[identity_key]['deployedOn'].append(agent['chainId'])
+
+        logger.info(
+            f"Deduplication: {len(agents)} agents → {len(deduplicated)} unique agents"
+        )
+
+        return deduplicated
+
+    def _sort_agents_cross_chain(
+        self,
+        agents: List[Dict[str, Any]],
+        sort: List[str]
+    ) -> List[Dict[str, Any]]:
+        """
+        Sort agents from multiple chains.
+
+        Supports sorting by:
+        - createdAt (timestamp)
+        - updatedAt (timestamp)
+        - totalFeedback (count)
+        - name (alphabetical)
+        - averageScore (reputation, if available)
+        """
+        if not sort or len(sort) == 0:
+            # Default: sort by createdAt descending (newest first)
+            return sorted(
+                agents,
+                key=lambda a: a.get('createdAt', 0),
+                reverse=True
+            )
+
+        # Parse first sort specification
+        sort_spec = sort[0]
+        if ':' in sort_spec:
+            field, direction = sort_spec.split(':', 1)
+        else:
+            field = sort_spec
+            direction = 'desc'
+
+        reverse = (direction.lower() == 'desc')
+
+        # Define sort key function
+        def get_sort_key(agent: Dict[str, Any]):
+            if field == 'createdAt':
+                return agent.get('createdAt', 0)
+
+            elif field == 'updatedAt':
+                return agent.get('updatedAt', 0)
+
+            elif field == 'totalFeedback':
+                return agent.get('totalFeedback', 0)
+
+            elif field == 'name':
+                reg_file = agent.get('registrationFile', {})
+                return reg_file.get('name', '').lower()
+
+            elif field == 'averageScore':
+                # If reputation search was done, averageScore may be available
+                return agent.get('averageScore', 0)
+
+            else:
+                logger.warning(f"Unknown sort field: {field}, defaulting to createdAt")
+                return agent.get('createdAt', 0)
+
+        return sorted(agents, key=get_sort_key, reverse=reverse)
+
+    def _parse_multi_chain_cursor(self, cursor: Optional[str]) -> Dict[int, int]:
+        """
+        Parse multi-chain cursor into per-chain offsets.
+
+        Cursor format (JSON):
+        {
+            "11155111": 50,  # Ethereum Sepolia offset
+            "84532": 30,     # Base Sepolia offset
+            "_global_offset": 100  # Total items returned so far
+        }
+
+        Returns:
+            Dict mapping chainId → offset (default 0)
+        """
+        if not cursor:
+            return {}
+
+        try:
+            cursor_data = json.loads(cursor)
+
+            # Validate format
+            if not isinstance(cursor_data, dict):
+                logger.warning(f"Invalid cursor format: {cursor}, using empty")
+                return {}
+
+            return cursor_data
+
+        except json.JSONDecodeError as e:
+            logger.warning(f"Failed to parse cursor: {e}, using empty")
+            return {}
+
+    def _create_multi_chain_cursor(
+        self,
+        global_offset: int,
+    ) -> str:
+        """
+        Create multi-chain cursor for next page.
+
+        Args:
+            global_offset: Total items returned so far
+
+        Returns:
+            JSON string cursor
+        """
+        cursor_data = {
+            "_global_offset": global_offset
+        }
+
+        return json.dumps(cursor_data)
+
+    def _extract_order_by(self, sort: List[str]) -> str:
+        """Extract order_by field from sort specification."""
+        if not sort or len(sort) == 0:
+            return "createdAt"
+
+        sort_spec = sort[0]
+        if ':' in sort_spec:
+            field, _ = sort_spec.split(':', 1)
+            return field
+        return sort_spec
+
+    def _extract_order_direction(self, sort: List[str]) -> str:
+        """Extract order direction from sort specification."""
+        if not sort or len(sort) == 0:
+            return "desc"
+
+        sort_spec = sort[0]
+        if ':' in sort_spec:
+            _, direction = sort_spec.split(':', 1)
+            return direction
+        return "desc"