feat: Settings page, DELETE traces endpoint, Anthropic SDK, dashboard bug fixes

- Add /dashboard/settings with SDK connection details, data stats, purge
- Add DELETE /api/traces/[id] with cascade deletion
- Add Anthropic integration (wrap_anthropic) for Python SDK
- Fix missing root duration (totalDuration -> durationMs mapping)
- Fix truncated JSON in decision tree nodes (extract readable labels)
- Fix hardcoded 128K maxTokens in token gauge (model-aware context windows)
- Enable Settings nav item in sidebar

apps/web/src/app/api/settings/purge/route.ts

@@ -0,0 +1,21 @@
+import { NextResponse } from "next/server";
+import { prisma } from "@/lib/prisma";
+
+export async function POST() {
+  try {
+    await prisma.$transaction([
+      prisma.event.deleteMany(),
+      prisma.decisionPoint.deleteMany(),
+      prisma.span.deleteMany(),
+      prisma.trace.deleteMany(),
+    ]);
+
+    return NextResponse.json({ success: true }, { status: 200 });
+  } catch (error) {
+    console.error("Error purging data:", error);
+    return NextResponse.json(
+      { error: "Internal server error" },
+      { status: 500 }
+    );
+  }
+}

apps/web/src/app/api/settings/stats/route.ts

@@ -0,0 +1,25 @@
+import { NextResponse } from "next/server";
+import { prisma } from "@/lib/prisma";
+
+export async function GET() {
+  try {
+    const [totalTraces, totalSpans, totalDecisions, totalEvents] =
+      await Promise.all([
+        prisma.trace.count(),
+        prisma.span.count(),
+        prisma.decisionPoint.count(),
+        prisma.event.count(),
+      ]);
+
+    return NextResponse.json(
+      { totalTraces, totalSpans, totalDecisions, totalEvents },
+      { status: 200 }
+    );
+  } catch (error) {
+    console.error("Error fetching stats:", error);
+    return NextResponse.json(
+      { error: "Internal server error" },
+      { status: 500 }
+    );
+  }
+}

apps/web/src/app/api/traces/[id]/route.ts

@@ -1,10 +1,26 @@
-import { NextResponse } from "next/server";
+import { NextRequest, NextResponse } from "next/server";
 import { prisma } from "@/lib/prisma";
 
+type RouteParams = { params: Promise<{ id: string }> };
+
+function extractActionLabel(value: unknown): string {
+  if (typeof value === "string") return value;
+  if (value && typeof value === "object" && !Array.isArray(value)) {
+    const obj = value as Record<string, unknown>;
+    if (typeof obj.name === "string") return obj.name;
+    if (typeof obj.action === "string") return obj.action;
+    if (typeof obj.tool === "string") return obj.tool;
+    for (const v of Object.values(obj)) {
+      if (typeof v === "string") return v;
+    }
+  }
+  return JSON.stringify(value);
+}
+
 // GET /api/traces/[id] — Get single trace with all relations
 export async function GET(
-  _request: Request,
-  { params }: { params: Promise<{ id: string }> }
+  _request: NextRequest,
+  { params }: RouteParams
 ) {
   try {
     const { id } = await params;
@@ -41,11 +57,13 @@ export async function GET(
     // Transform data to match frontend expectations
     const transformedTrace = {
       ...trace,
+      durationMs: trace.totalDuration,
+      costUsd: trace.totalCost,
       decisionPoints: trace.decisionPoints.map((dp) => ({
         id: dp.id,
         type: dp.type,
-        chosenAction: typeof dp.chosen === "string" ? dp.chosen : JSON.stringify(dp.chosen),
-        alternatives: dp.alternatives.map((alt) => (typeof alt === "string" ? alt : JSON.stringify(alt))),
+        chosenAction: extractActionLabel(dp.chosen),
+        alternatives: dp.alternatives.map((alt) => extractActionLabel(alt)),
         reasoning: dp.reasoning,
         contextSnapshot: dp.contextSnapshot as Record<string, unknown> | null,
         confidence: null, // Not in schema, default to null
@@ -81,3 +99,35 @@ export async function GET(
     return NextResponse.json({ error: "Internal server error" }, { status: 500 });
   }
 }
+
+// DELETE /api/traces/[id] — Delete a trace and all related data (cascade)
+export async function DELETE(
+  _request: NextRequest,
+  { params }: RouteParams
+) {
+  try {
+    const { id } = await params;
+
+    if (!id || typeof id !== "string") {
+      return NextResponse.json({ error: "Invalid trace ID" }, { status: 400 });
+    }
+
+    const trace = await prisma.trace.findUnique({
+      where: { id },
+      select: { id: true },
+    });
+
+    if (!trace) {
+      return NextResponse.json({ error: "Trace not found" }, { status: 404 });
+    }
+
+    await prisma.trace.delete({
+      where: { id },
+    });
+
+    return NextResponse.json({ success: true, deleted: id }, { status: 200 });
+  } catch (error) {
+    console.error("Error deleting trace:", error);
+    return NextResponse.json({ error: "Internal server error" }, { status: 500 });
+  }
+}

apps/web/src/app/dashboard/layout.tsx

@@ -22,7 +22,7 @@ interface NavItem {
 const navItems: NavItem[] = [
   { href: "/dashboard", label: "Traces", icon: Activity },
   { href: "/dashboard/decisions", label: "Decisions", icon: GitBranch },
-  { href: "/dashboard/settings", label: "Settings", icon: Settings, comingSoon: true },
+  { href: "/dashboard/settings", label: "Settings", icon: Settings },
 ];
 
 function Sidebar({ onNavigate }: { onNavigate?: () => void }) {

apps/web/src/app/dashboard/settings/page.tsx

@@ -0,0 +1,294 @@
+"use client";
+
+import { useState, useEffect, useCallback } from "react";
+import {
+  Settings,
+  Key,
+  Globe,
+  Copy,
+  Check,
+  RefreshCw,
+  Database,
+  Trash2,
+  AlertTriangle,
+} from "lucide-react";
+import { cn } from "@/lib/utils";
+
+interface Stats {
+  totalTraces: number;
+  totalSpans: number;
+  totalDecisions: number;
+  totalEvents: number;
+}
+
+export default function SettingsPage() {
+  const [stats, setStats] = useState<Stats | null>(null);
+  const [isLoadingStats, setIsLoadingStats] = useState(true);
+  const [copiedField, setCopiedField] = useState<string | null>(null);
+  const [isPurging, setIsPurging] = useState(false);
+  const [showPurgeConfirm, setShowPurgeConfirm] = useState(false);
+
+  const fetchStats = useCallback(async () => {
+    setIsLoadingStats(true);
+    try {
+      const res = await fetch("/api/settings/stats", { cache: "no-store" });
+      if (res.ok) {
+        const data = await res.json();
+        setStats(data);
+      }
+    } catch (error) {
+      console.error("Failed to fetch stats:", error);
+    } finally {
+      setIsLoadingStats(false);
+    }
+  }, []);
+
+  useEffect(() => {
+    fetchStats();
+  }, [fetchStats]);
+
+  const copyToClipboard = async (text: string, field: string) => {
+    try {
+      await navigator.clipboard.writeText(text);
+      setCopiedField(field);
+      setTimeout(() => setCopiedField(null), 2000);
+    } catch {
+      console.error("Failed to copy");
+    }
+  };
+
+  const handlePurgeAll = async () => {
+    setIsPurging(true);
+    try {
+      const res = await fetch("/api/settings/purge", { method: "POST" });
+      if (res.ok) {
+        setShowPurgeConfirm(false);
+        fetchStats();
+      }
+    } catch (error) {
+      console.error("Failed to purge:", error);
+    } finally {
+      setIsPurging(false);
+    }
+  };
+
+  const endpointUrl =
+    typeof window !== "undefined"
+      ? `${window.location.origin}/api/traces`
+      : "https://agentlens.vectry.tech/api/traces";
+
+  return (
+    <div className="space-y-8 max-w-3xl">
+      <div>
+        <h1 className="text-2xl font-bold text-neutral-100">Settings</h1>
+        <p className="text-neutral-400 mt-1">
+          Configuration and SDK connection details
+        </p>
+      </div>
+
+      {/* SDK Connection */}
+      <section className="space-y-4">
+        <div className="flex items-center gap-2 text-neutral-300">
+          <Globe className="w-5 h-5 text-emerald-400" />
+          <h2 className="text-lg font-semibold">SDK Connection</h2>
+        </div>
+
+        <div className="bg-neutral-900 border border-neutral-800 rounded-xl p-6 space-y-5">
+          <SettingField
+            label="Ingest Endpoint"
+            value={endpointUrl}
+            copiedField={copiedField}
+            fieldKey="endpoint"
+            onCopy={copyToClipboard}
+          />
+
+          <SettingField
+            label="API Key"
+            value="any-value-accepted"
+            hint="Authentication is not enforced yet. Use any non-empty string as your Bearer token."
+            copiedField={copiedField}
+            fieldKey="apikey"
+            onCopy={copyToClipboard}
+          />
+
+          <div className="pt-4 border-t border-neutral-800">
+            <p className="text-xs text-neutral-500 mb-3">Quick start</p>
+            <div className="bg-neutral-950 border border-neutral-800 rounded-lg p-4 font-mono text-sm text-neutral-300 overflow-x-auto">
+              <pre>{`from agentlens import init
+
+init(
+    api_key="your-api-key",
+    endpoint="${endpointUrl.replace("/api/traces", "")}",
+)`}</pre>
+            </div>
+          </div>
+        </div>
+      </section>
+
+      {/* Data & Storage */}
+      <section className="space-y-4">
+        <div className="flex items-center gap-2 text-neutral-300">
+          <Database className="w-5 h-5 text-emerald-400" />
+          <h2 className="text-lg font-semibold">Data & Storage</h2>
+        </div>
+
+        <div className="bg-neutral-900 border border-neutral-800 rounded-xl p-6 space-y-5">
+          {isLoadingStats ? (
+            <div className="grid grid-cols-2 sm:grid-cols-4 gap-4">
+              {Array.from({ length: 4 }).map((_, i) => (
+                <div key={i} className="animate-pulse">
+                  <div className="h-4 w-16 bg-neutral-800 rounded mb-2" />
+                  <div className="h-8 w-12 bg-neutral-800 rounded" />
+                </div>
+              ))}
+            </div>
+          ) : stats ? (
+            <div className="grid grid-cols-2 sm:grid-cols-4 gap-4">
+              <StatCard label="Traces" value={stats.totalTraces} />
+              <StatCard label="Spans" value={stats.totalSpans} />
+              <StatCard label="Decisions" value={stats.totalDecisions} />
+              <StatCard label="Events" value={stats.totalEvents} />
+            </div>
+          ) : (
+            <p className="text-sm text-neutral-500">
+              Unable to load statistics
+            </p>
+          )}
+
+          <div className="pt-4 border-t border-neutral-800 flex items-center justify-between">
+            <div>
+              <p className="text-sm text-neutral-300 font-medium">
+                Purge All Data
+              </p>
+              <p className="text-xs text-neutral-500 mt-0.5">
+                Permanently delete all traces, spans, decisions, and events
+              </p>
+            </div>
+            {showPurgeConfirm ? (
+              <div className="flex items-center gap-2">
+                <button
+                  onClick={() => setShowPurgeConfirm(false)}
+                  className="px-3 py-2 text-sm text-neutral-400 hover:text-neutral-200 transition-colors"
+                >
+                  Cancel
+                </button>
+                <button
+                  onClick={handlePurgeAll}
+                  disabled={isPurging}
+                  className="flex items-center gap-2 px-4 py-2 bg-red-500/20 border border-red-500/30 text-red-400 rounded-lg text-sm font-medium hover:bg-red-500/30 disabled:opacity-50 transition-colors"
+                >
+                  {isPurging ? (
+                    <RefreshCw className="w-4 h-4 animate-spin" />
+                  ) : (
+                    <AlertTriangle className="w-4 h-4" />
+                  )}
+                  Confirm Purge
+                </button>
+              </div>
+            ) : (
+              <button
+                onClick={() => setShowPurgeConfirm(true)}
+                className="flex items-center gap-2 px-4 py-2 bg-neutral-800 border border-neutral-700 text-neutral-400 rounded-lg text-sm font-medium hover:text-red-400 hover:border-red-500/30 transition-colors"
+              >
+                <Trash2 className="w-4 h-4" />
+                Purge
+              </button>
+            )}
+          </div>
+        </div>
+      </section>
+
+      {/* About */}
+      <section className="space-y-4">
+        <div className="flex items-center gap-2 text-neutral-300">
+          <Settings className="w-5 h-5 text-emerald-400" />
+          <h2 className="text-lg font-semibold">About</h2>
+        </div>
+
+        <div className="bg-neutral-900 border border-neutral-800 rounded-xl p-6">
+          <div className="grid grid-cols-2 gap-4 text-sm">
+            <div>
+              <p className="text-neutral-500">Version</p>
+              <p className="text-neutral-200 font-medium">0.1.0</p>
+            </div>
+            <div>
+              <p className="text-neutral-500">SDK Package</p>
+              <p className="text-neutral-200 font-medium">vectry-agentlens</p>
+            </div>
+            <div>
+              <p className="text-neutral-500">Database</p>
+              <p className="text-neutral-200 font-medium">PostgreSQL</p>
+            </div>
+            <div>
+              <p className="text-neutral-500">License</p>
+              <p className="text-neutral-200 font-medium">MIT</p>
+            </div>
+          </div>
+        </div>
+      </section>
+    </div>
+  );
+}
+
+function SettingField({
+  label,
+  value,
+  hint,
+  copiedField,
+  fieldKey,
+  onCopy,
+}: {
+  label: string;
+  value: string;
+  hint?: string;
+  copiedField: string | null;
+  fieldKey: string;
+  onCopy: (text: string, field: string) => void;
+}) {
+  const isCopied = copiedField === fieldKey;
+
+  return (
+    <div>
+      <label className="text-xs text-neutral-500 font-medium block mb-1.5">
+        {label}
+      </label>
+      <div className="flex items-center gap-2">
+        <div className="flex-1 flex items-center gap-2 px-3 py-2.5 bg-neutral-950 border border-neutral-800 rounded-lg">
+          <Key className="w-4 h-4 text-neutral-600 shrink-0" />
+          <code className="text-sm text-neutral-300 font-mono truncate">
+            {value}
+          </code>
+        </div>
+        <button
+          onClick={() => onCopy(value, fieldKey)}
+          className={cn(
+            "p-2.5 rounded-lg border transition-all",
+            isCopied
+              ? "bg-emerald-500/10 border-emerald-500/30 text-emerald-400"
+              : "bg-neutral-800 border-neutral-700 text-neutral-400 hover:text-neutral-200"
+          )}
+        >
+          {isCopied ? (
+            <Check className="w-4 h-4" />
+          ) : (
+            <Copy className="w-4 h-4" />
+          )}
+        </button>
+      </div>
+      {hint && (
+        <p className="text-xs text-neutral-600 mt-1.5">{hint}</p>
+      )}
+    </div>
+  );
+}
+
+function StatCard({ label, value }: { label: string; value: number }) {
+  return (
+    <div className="p-3 bg-neutral-800/50 rounded-lg">
+      <p className="text-xs text-neutral-500">{label}</p>
+      <p className="text-xl font-bold text-neutral-100 mt-1">
+        {value.toLocaleString()}
+      </p>
+    </div>
+  );
+}

apps/web/src/components/trace-analytics.tsx

@@ -461,13 +461,35 @@ function CostBreakdown({
 // Section C: Token Usage Gauge
 function TokenUsageGauge({ trace }: { trace: Trace }) {
   const tokenData = useMemo(() => {
-    // Try to get total tokens from various sources
     const totalTokens =
       (trace.metadata?.totalTokens as number | null | undefined) ??
       (trace.metadata?.tokenCount as number | null | undefined) ??
       null;
 
-    const maxTokens = 128000; // Default context window
+    const modelContextWindows: Record<string, number> = {
+      "gpt-4": 8192,
+      "gpt-4-32k": 32768,
+      "gpt-4-turbo": 128000,
+      "gpt-4o": 128000,
+      "gpt-4o-mini": 128000,
+      "gpt-3.5-turbo": 16385,
+      "claude-3-opus": 200000,
+      "claude-3-sonnet": 200000,
+      "claude-3-haiku": 200000,
+      "claude-3.5-sonnet": 200000,
+      "claude-4-opus": 200000,
+      "claude-4-sonnet": 200000,
+    };
+
+    const model = (trace.metadata?.model as string | undefined) ?? "";
+    const modelLower = model.toLowerCase();
+    let maxTokens = 128000;
+    for (const [prefix, ctx] of Object.entries(modelContextWindows)) {
+      if (modelLower.startsWith(prefix)) {
+        maxTokens = ctx;
+        break;
+      }
+    }
 
     return {
       totalTokens,

packages/sdk-python/agentlens/integrations/__init__.py

@@ -1 +1,8 @@
-"""Integration packages for AgentLens."""
+"""Integration packages for AgentLens.
+
+Available integrations:
+
+- ``openai``: Wrap OpenAI clients with ``wrap_openai(client)``.
+- ``anthropic``: Wrap Anthropic clients with ``wrap_anthropic(client)``.
+- ``langchain``: LangChain callback handler for tracing.
+"""

packages/sdk-python/agentlens/integrations/anthropic.py

@@ -0,0 +1,697 @@
+"""Anthropic integration for AgentLens.
+
+This module provides a wrapper that auto-instruments Anthropic API calls with
+tracing, span creation, decision logging for tool calls, and token tracking.
+"""
+
+import json
+import logging
+import time
+from functools import wraps
+from typing import Any, Dict, Iterator, List, Optional
+
+from agentlens.models import (
+    Event,
+    EventType,
+    _now_iso,
+)
+from agentlens.trace import (
+    TraceContext,
+    _get_context_stack,
+    get_current_span_id,
+    get_current_trace,
+)
+
+logger = logging.getLogger("agentlens")
+
+# Cost per 1K tokens (input/output) for common Claude models
+_MODEL_COSTS: Dict[str, tuple] = {
+    # Claude 3 family
+    "claude-3-opus-20240229": (0.015, 0.075),
+    "claude-3-sonnet-20240229": (0.003, 0.015),
+    "claude-3-haiku-20240307": (0.00025, 0.00125),
+    # Claude 3.5 family
+    "claude-3-5-sonnet-20240620": (0.003, 0.015),
+    "claude-3-5-sonnet-20241022": (0.003, 0.015),
+    "claude-3-5-haiku-20241022": (0.0008, 0.004),
+    # Claude 4 family
+    "claude-sonnet-4-20250514": (0.003, 0.015),
+    "claude-opus-4-20250514": (0.015, 0.075),
+    # Short aliases for prefix matching
+    "claude-3-opus": (0.015, 0.075),
+    "claude-3-sonnet": (0.003, 0.015),
+    "claude-3-haiku": (0.00025, 0.00125),
+    "claude-3-5-sonnet": (0.003, 0.015),
+    "claude-3-5-haiku": (0.0008, 0.004),
+    "claude-3.5-sonnet": (0.003, 0.015),
+    "claude-3.5-haiku": (0.0008, 0.004),
+    "claude-sonnet-4": (0.003, 0.015),
+    "claude-opus-4": (0.015, 0.075),
+    "claude-4-sonnet": (0.003, 0.015),
+    "claude-4-opus": (0.015, 0.075),
+}
+
+
+def _truncate_data(data: Any, max_length: int = 500) -> Any:
+    """Truncate data for privacy while preserving structure."""
+    if isinstance(data, str):
+        return data[:max_length] + "..." if len(data) > max_length else data
+    elif isinstance(data, dict):
+        return {k: _truncate_data(v, max_length) for k, v in data.items()}
+    elif isinstance(data, list):
+        return [_truncate_data(item, max_length) for item in data]
+    else:
+        return data
+
+
+def _calculate_cost(
+    model: str, input_tokens: int, output_tokens: int
+) -> Optional[float]:
+    """Calculate cost in USD based on model pricing."""
+    model_lower = model.lower()
+
+    if model_lower in _MODEL_COSTS:
+        input_cost, output_cost = _MODEL_COSTS[model_lower]
+        return (float(input_tokens) / 1000.0) * input_cost + float(
+            output_tokens
+        ) / 1000.0 * output_cost
+
+    best_match = None
+    best_len = 0
+    for model_name, costs in _MODEL_COSTS.items():
+        if model_lower.startswith(model_name.lower()) and len(model_name) > best_len:
+            best_match = costs
+            best_len = len(model_name)
+
+    if best_match:
+        input_cost, output_cost = best_match
+        return (float(input_tokens) / 1000.0) * input_cost + float(
+            output_tokens
+        ) / 1000.0 * output_cost
+
+    return None
+
+
+def _extract_messages_truncated(messages: List[Any]) -> List[Dict[str, Any]]:
+    """Extract and truncate message content."""
+    truncated = []
+    for msg in messages:
+        if isinstance(msg, dict):
+            truncated_msg = {"role": msg.get("role", "unknown")}
+            content = msg.get("content")
+            if content is not None:
+                if isinstance(content, list):
+                    # Anthropic supports content as list of blocks
+                    truncated_msg["content"] = _truncate_data(content)
+                else:
+                    truncated_msg["content"] = _truncate_data(str(content))
+            truncated.append(truncated_msg)
+        else:
+            # Handle message objects
+            role = getattr(msg, "role", "unknown")
+            content = getattr(msg, "content", "")
+            truncated.append({"role": role, "content": _truncate_data(str(content))})
+    return truncated
+
+
+def _extract_content_from_response(response: Any) -> Optional[str]:
+    """Extract text content from Anthropic response.
+
+    Anthropic responses have a ``content`` array with blocks of type
+    ``text`` or ``tool_use``.
+    """
+    if hasattr(response, "content") and response.content:
+        text_parts = []
+        for block in response.content:
+            if hasattr(block, "type") and block.type == "text":
+                text_parts.append(getattr(block, "text", ""))
+            elif isinstance(block, dict) and block.get("type") == "text":
+                text_parts.append(block.get("text", ""))
+        if text_parts:
+            return _truncate_data(" ".join(text_parts))
+    return None
+
+
+def _extract_tool_calls_from_response(response: Any) -> List[Dict[str, Any]]:
+    """Extract tool_use blocks from Anthropic response.
+
+    Anthropic tool calls appear as content blocks with ``type: "tool_use"``,
+    containing ``name`` and ``input`` fields.
+    """
+    tool_calls: List[Dict[str, Any]] = []
+    if hasattr(response, "content") and response.content:
+        for block in response.content:
+            block_type = getattr(block, "type", None) or (
+                block.get("type") if isinstance(block, dict) else None
+            )
+            if block_type == "tool_use":
+                if isinstance(block, dict):
+                    name = block.get("name", "unknown")
+                    arguments = block.get("input", {})
+                else:
+                    name = getattr(block, "name", "unknown")
+                    arguments = getattr(block, "input", {})
+                tool_calls.append({"name": name, "arguments": arguments})
+    return tool_calls
+
+
+class _StreamWrapper:
+    """Wrapper for Anthropic stream responses to collect events and finalize span."""
+
+    def __init__(self, original_stream: Any, trace_ctx: Optional[TraceContext]):
+        self._original_stream = original_stream
+        self._trace_ctx = trace_ctx
+        self._events: List[Any] = []
+        self._start_time = time.time()
+        self._model: Optional[str] = None
+        self._temperature: Optional[float] = None
+        self._max_tokens: Optional[int] = None
+        self._messages: Optional[List[Any]] = None
+        self._parent_span_id = get_current_span_id()
+        # Accumulated response data from stream events
+        self._text_content: str = ""
+        self._tool_calls: List[Dict[str, Any]] = []
+        self._current_tool: Optional[Dict[str, Any]] = None
+        self._input_tokens: Optional[int] = None
+        self._output_tokens: Optional[int] = None
+        self._response_model: Optional[str] = None
+        self._stop_reason: Optional[str] = None
+
+    def set_params(
+        self,
+        model: str,
+        temperature: Optional[float],
+        max_tokens: Optional[int],
+        messages: List[Any],
+    ) -> None:
+        self._model = model
+        self._temperature = temperature
+        self._max_tokens = max_tokens
+        self._messages = messages
+
+    def _process_event(self, event: Any) -> None:
+        """Process a single stream event to accumulate response data."""
+        event_type = getattr(event, "type", None)
+
+        if event_type == "message_start":
+            message = getattr(event, "message", None)
+            if message:
+                self._response_model = getattr(message, "model", None)
+                usage = getattr(message, "usage", None)
+                if usage:
+                    self._input_tokens = getattr(usage, "input_tokens", None)
+
+        elif event_type == "content_block_start":
+            block = getattr(event, "content_block", None)
+            if block:
+                block_type = getattr(block, "type", None)
+                if block_type == "tool_use":
+                    self._current_tool = {
+                        "name": getattr(block, "name", "unknown"),
+                        "arguments": "",
+                    }
+
+        elif event_type == "content_block_delta":
+            delta = getattr(event, "delta", None)
+            if delta:
+                delta_type = getattr(delta, "type", None)
+                if delta_type == "text_delta":
+                    self._text_content += getattr(delta, "text", "")
+                elif delta_type == "input_json_delta":
+                    if self._current_tool is not None:
+                        self._current_tool["arguments"] += getattr(
+                            delta, "partial_json", ""
+                        )
+
+        elif event_type == "content_block_stop":
+            if self._current_tool is not None:
+                # Parse accumulated JSON arguments
+                try:
+                    args_str = self._current_tool["arguments"]
+                    if isinstance(args_str, str) and args_str:
+                        self._current_tool["arguments"] = json.loads(args_str)
+                    elif not args_str:
+                        self._current_tool["arguments"] = {}
+                except (json.JSONDecodeError, TypeError):
+                    pass
+                self._tool_calls.append(self._current_tool)
+                self._current_tool = None
+
+        elif event_type == "message_delta":
+            delta = getattr(event, "delta", None)
+            if delta:
+                self._stop_reason = getattr(delta, "stop_reason", None)
+            usage = getattr(event, "usage", None)
+            if usage:
+                self._output_tokens = getattr(usage, "output_tokens", None)
+
+    def __iter__(self) -> Iterator[Any]:
+        return self
+
+    def __next__(self) -> Any:
+        event = next(self._original_stream)
+        self._events.append(event)
+        self._process_event(event)
+        return event
+
+    def __enter__(self) -> "_StreamWrapper":
+        """Support context manager protocol for Anthropic streaming."""
+        if hasattr(self._original_stream, "__enter__"):
+            self._original_stream.__enter__()
+        return self
+
+    def __exit__(
+        self,
+        exc_type: Optional[type],
+        exc_val: Optional[BaseException],
+        exc_tb: Optional[Any],
+    ) -> None:
+        """Finalize span and close underlying stream on context manager exit."""
+        if hasattr(self._original_stream, "__exit__"):
+            self._original_stream.__exit__(exc_type, exc_val, exc_tb)
+        self.finalize()
+
+    def finalize(self) -> None:
+        """Create span after stream is fully consumed."""
+        if not self._events:
+            return
+
+        response_model = self._response_model or self._model or "unknown"
+
+        # Build a mock response object for _create_llm_span
+        mock = _MockResponse()
+        mock.model = response_model
+        mock.text_content = self._text_content or None
+        mock.tool_calls = self._tool_calls
+        mock.stop_reason = self._stop_reason
+        mock.input_tokens = self._input_tokens
+        mock.output_tokens = self._output_tokens
+
+        _create_llm_span(
+            response=mock,
+            start_time=self._start_time,
+            model=self._model or response_model,
+            temperature=self._temperature,
+            max_tokens=self._max_tokens,
+            messages=self._messages or [],
+            parent_span_id=self._parent_span_id,
+            trace_ctx=self._trace_ctx,
+        )
+
+        if self._trace_ctx:
+            self._trace_ctx.__exit__(None, None, None)
+
+
+class _MockResponse:
+    """Lightweight object to unify stream-assembled and regular responses."""
+
+    def __init__(self) -> None:
+        self.model: str = "unknown"
+        self.text_content: Optional[str] = None
+        self.tool_calls: List[Dict[str, Any]] = []
+        self.stop_reason: Optional[str] = None
+        self.input_tokens: Optional[int] = None
+        self.output_tokens: Optional[int] = None
+        # Fake content list for compatibility with extraction helpers
+        self.content: List[Any] = []
+
+
+def _create_llm_span(
+    response: Any,
+    start_time: float,
+    model: str,
+    temperature: Optional[float],
+    max_tokens: Optional[int],
+    messages: List[Any],
+    parent_span_id: Optional[str],
+    trace_ctx: Optional[TraceContext],
+) -> None:
+    """Create LLM span from Anthropic response."""
+    from agentlens.models import Span, SpanStatus, SpanType
+
+    current_trace = get_current_trace()
+    if current_trace is None:
+        logger.warning("No active trace, skipping span creation")
+        return
+
+    end_time = time.time()
+    duration_ms = int((end_time - start_time) * 1000)
+
+    # Extract token usage
+    token_count = None
+    cost_usd = None
+
+    # Handle real Anthropic response
+    input_tokens = getattr(response, "input_tokens", None)
+    output_tokens = getattr(response, "output_tokens", None)
+
+    # Real responses have usage object
+    if input_tokens is None and hasattr(response, "usage"):
+        usage = response.usage
+        input_tokens = getattr(usage, "input_tokens", None)
+        output_tokens = getattr(usage, "output_tokens", None)
+
+    if input_tokens is not None and output_tokens is not None:
+        token_count = input_tokens + output_tokens
+        cost_usd = _calculate_cost(model, input_tokens, output_tokens)
+
+    # Extract content - try helpers first, fall back to mock fields
+    content = _extract_content_from_response(response)
+    if content is None:
+        text_content = getattr(response, "text_content", None)
+        if text_content:
+            content = _truncate_data(str(text_content))
+
+    # Extract tool calls - try helpers first, fall back to mock fields
+    tool_calls = _extract_tool_calls_from_response(response)
+    if not tool_calls:
+        tool_calls = getattr(response, "tool_calls", []) or []
+
+    # Extract stop reason
+    stop_reason = getattr(response, "stop_reason", None)
+
+    # Create span
+    span_name = f"anthropic.{model}"
+    span = Span(
+        name=span_name,
+        type=SpanType.LLM_CALL.value,
+        parent_span_id=parent_span_id,
+        input_data={"messages": _extract_messages_truncated(messages)},
+        output_data={"content": content, "tool_calls": tool_calls or None},
+        token_count=token_count,
+        cost_usd=cost_usd,
+        duration_ms=duration_ms,
+        status=SpanStatus.COMPLETED.value,
+        started_at=_now_iso(),
+        ended_at=_now_iso(),
+        metadata={
+            "model": model,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+            "stop_reason": stop_reason,
+        },
+    )
+
+    current_trace.spans.append(span)
+
+    # Push onto context stack for decision logging
+    stack = _get_context_stack()
+    stack.append(span)
+
+    # Log tool call decisions
+    if tool_calls:
+        from agentlens.decision import log_decision
+
+        # Try to get reasoning from the assistant's text content
+        reasoning = None
+        if content:
+            reasoning = _truncate_data(str(content))
+
+        # Build context snapshot
+        context_snapshot = None
+        if input_tokens is not None or output_tokens is not None:
+            context_snapshot = {
+                "model": model,
+                "input_tokens": input_tokens,
+                "output_tokens": output_tokens,
+            }
+
+        for tool_call in tool_calls:
+            log_decision(
+                type="TOOL_SELECTION",
+                chosen={
+                    "name": tool_call.get("name", "unknown"),
+                    "arguments": tool_call.get("arguments", {}),
+                },
+                alternatives=[],
+                reasoning=reasoning,
+                context_snapshot=context_snapshot,
+            )
+
+    # Always pop from context stack
+    if stack and stack[-1] == span:
+        stack.pop()
+    elif stack and isinstance(stack[-1], Span) and stack[-1].id == span.id:
+        stack.pop()
+
+
+def _handle_error(
+    error: Exception,
+    start_time: float,
+    model: str,
+    temperature: Optional[float],
+    max_tokens: Optional[int],
+    messages: List[Any],
+    parent_span_id: Optional[str],
+    trace_ctx: Optional[TraceContext],
+) -> None:
+    """Handle error by creating error span and event."""
+    from agentlens.models import Span, SpanStatus, SpanType
+
+    current_trace = get_current_trace()
+    if current_trace is None:
+        return
+
+    end_time = time.time()
+    duration_ms = int((end_time - start_time) * 1000)
+
+    # Create error span
+    span_name = f"anthropic.{model}"
+    span = Span(
+        name=span_name,
+        type=SpanType.LLM_CALL.value,
+        parent_span_id=parent_span_id,
+        input_data={"messages": _extract_messages_truncated(messages)},
+        status=SpanStatus.ERROR.value,
+        status_message=str(error),
+        started_at=_now_iso(),
+        ended_at=_now_iso(),
+        duration_ms=duration_ms,
+        metadata={
+            "model": model,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+        },
+    )
+
+    current_trace.spans.append(span)
+
+    # Create error event
+    error_event = Event(
+        type=EventType.ERROR.value,
+        name=f"{span_name}: {str(error)}",
+        span_id=span.id,
+        metadata={"error_type": type(error).__name__},
+    )
+
+    current_trace.events.append(error_event)
+
+    # Pop from context stack if needed
+    stack = _get_context_stack()
+    if stack and isinstance(stack[-1], Span) and stack[-1].id == span.id:
+        stack.pop()
+
+
+def _wrap_create(original_create: Any, is_async: bool = False) -> Any:
+    """Wrap Anthropic messages.create method."""
+
+    if is_async:
+
+        @wraps(original_create)
+        async def async_traced_create(*args: Any, **kwargs: Any) -> Any:
+            # Extract parameters
+            model = kwargs.get("model", "claude-3-5-sonnet-20241022")
+            temperature = kwargs.get("temperature")
+            max_tokens = kwargs.get("max_tokens")
+            messages = kwargs.get("messages", [])
+            stream = kwargs.get("stream", False)
+
+            parent_span_id = get_current_span_id()
+            start_time = time.time()
+
+            # Handle streaming
+            if stream:
+                trace_ctx = None
+                if get_current_trace() is None:
+                    trace_ctx = TraceContext(name=f"anthropic-{model}")
+                    trace_ctx.__enter__()
+
+                try:
+                    original_stream = await original_create(*args, **kwargs)
+
+                    wrapper = _StreamWrapper(original_stream, trace_ctx)
+                    wrapper.set_params(model, temperature, max_tokens, messages)
+
+                    return wrapper
+                except Exception as e:
+                    if trace_ctx:
+                        trace_ctx.__exit__(type(e), e, None)
+                    raise
+
+            # Non-streaming
+            trace_ctx = None
+            if get_current_trace() is None:
+                trace_ctx = TraceContext(name=f"anthropic-{model}")
+                trace_ctx.__enter__()
+
+            try:
+                response = await original_create(*args, **kwargs)
+
+                _create_llm_span(
+                    response=response,
+                    start_time=start_time,
+                    model=model,
+                    temperature=temperature,
+                    max_tokens=max_tokens,
+                    messages=messages,
+                    parent_span_id=parent_span_id,
+                    trace_ctx=trace_ctx,
+                )
+
+                if trace_ctx is not None:
+                    trace_ctx.__exit__(None, None, None)
+
+                return response
+            except Exception as e:
+                _handle_error(
+                    error=e,
+                    start_time=start_time,
+                    model=model,
+                    temperature=temperature,
+                    max_tokens=max_tokens,
+                    messages=messages,
+                    parent_span_id=parent_span_id,
+                    trace_ctx=trace_ctx,
+                )
+                raise
+
+        return async_traced_create
+
+    else:
+
+        @wraps(original_create)
+        def traced_create(*args: Any, **kwargs: Any) -> Any:
+            # Extract parameters
+            model = kwargs.get("model", "claude-3-5-sonnet-20241022")
+            temperature = kwargs.get("temperature")
+            max_tokens = kwargs.get("max_tokens")
+            messages = kwargs.get("messages", [])
+            stream = kwargs.get("stream", False)
+
+            parent_span_id = get_current_span_id()
+            start_time = time.time()
+
+            # Handle streaming
+            if stream:
+                trace_ctx = None
+                if get_current_trace() is None:
+                    trace_ctx = TraceContext(name=f"anthropic-{model}")
+                    trace_ctx.__enter__()
+
+                try:
+                    original_stream = original_create(*args, **kwargs)
+
+                    wrapper = _StreamWrapper(original_stream, trace_ctx)
+                    wrapper.set_params(model, temperature, max_tokens, messages)
+
+                    return wrapper
+                except Exception as e:
+                    if trace_ctx:
+                        trace_ctx.__exit__(type(e), e, None)
+                    raise
+
+            # Non-streaming
+            trace_ctx = None
+            if get_current_trace() is None:
+                trace_ctx = TraceContext(name=f"anthropic-{model}")
+                trace_ctx.__enter__()
+
+            try:
+                response = original_create(*args, **kwargs)
+
+                _create_llm_span(
+                    response=response,
+                    start_time=start_time,
+                    model=model,
+                    temperature=temperature,
+                    max_tokens=max_tokens,
+                    messages=messages,
+                    parent_span_id=parent_span_id,
+                    trace_ctx=trace_ctx,
+                )
+
+                if trace_ctx is not None:
+                    trace_ctx.__exit__(None, None, None)
+
+                return response
+            except Exception as e:
+                _handle_error(
+                    error=e,
+                    start_time=start_time,
+                    model=model,
+                    temperature=temperature,
+                    max_tokens=max_tokens,
+                    messages=messages,
+                    parent_span_id=parent_span_id,
+                    trace_ctx=trace_ctx,
+                )
+                raise
+
+        return traced_create
+
+
+def wrap_anthropic(client: Any) -> Any:
+    """Wrap an Anthropic client to add AgentLens tracing.
+
+    Instruments ``client.messages.create()`` to automatically capture LLM spans,
+    token usage, cost estimation, and tool-call decisions.
+
+    Supports both sync (``anthropic.Anthropic``) and async
+    (``anthropic.AsyncAnthropic``) clients as well as streaming responses.
+
+    Args:
+        client: An ``anthropic.Anthropic`` or ``anthropic.AsyncAnthropic`` instance.
+
+    Returns:
+        The same client instance with ``messages.create`` wrapped for tracing.
+
+    Example::
+
+        import anthropic
+        from agentlens.integrations.anthropic import wrap_anthropic
+
+        client = anthropic.Anthropic(api_key="sk-...")
+        traced_client = wrap_anthropic(client)
+
+        response = traced_client.messages.create(
+            model="claude-3-sonnet-20240229",
+            max_tokens=1024,
+            messages=[{"role": "user", "content": "Hello!"}]
+        )
+    """
+    # Detect async client by checking for common async patterns
+    is_async = False
+    try:
+        import asyncio
+        import inspect
+
+        create_method = client.messages.create
+        if inspect.iscoroutinefunction(create_method) or (
+            hasattr(create_method, "__wrapped__")
+            and inspect.iscoroutinefunction(create_method.__wrapped__)
+        ):
+            is_async = True
+    except (AttributeError, ImportError):
+        pass
+
+    # Also detect by class name as a fallback
+    client_class_name = type(client).__name__
+    if "Async" in client_class_name:
+        is_async = True
+
+    original_create = client.messages.create
+    traced_create = _wrap_create(original_create, is_async=is_async)
+    client.messages.create = traced_create
+
+    logger.debug("Anthropic client wrapped with AgentLens tracing")
+    return client