diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/CMakeLists.txt b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b7fe9dda55c3375cefde6fb7c0e363d1b32d2714
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/CMakeLists.txt
@@ -0,0 +1,69 @@
+cmake_minimum_required(VERSION 3.5.1)
+project(fuse_matmul_add_pass)
+
+set(CMAKE_SKIP_INSTALL_ALL_DEPENDENCY TRUE)
+
+if (DEFINED ENV{ASCEND_INSTALL_PATH})
+    set(ASCEND_PATH $ENV{ASCEND_INSTALL_PATH})
+else ()
+    set(ASCEND_PATH /home/hugo/ascend_pkg/toolkit/latest)
+endif()
+
+set(INCLUDE_DIR ${ASCEND_PATH}/include)
+aux_source_directory(${PROJECT_SOURCE_DIR}/src SRC_LIST)
+
+set(common_compile_options
+    -std=c++17
+    -g
+    -Wall
+)
+
+set(common_compile_definitions
+    _GLIBCXX_USE_CXX11_ABI=0
+)
+message(STATUS "ASCEND_PATH: ${ASCEND_PATH}")
+
+add_subdirectory(gen_es_api)
+
+############ libfuse_matmul_add_pass.so ############
+add_library(${PROJECT_NAME} SHARED ${SRC_LIST})
+
+set_target_properties(${PROJECT_NAME} PROPERTIES output_name ${PROJECT_NAME})
+
+target_compile_options(${PROJECT_NAME} PRIVATE
+    ${common_compile_options}
+
+)
+
+target_compile_definitions(${PROJECT_NAME} PRIVATE
+    ${common_compile_definitions}
+)
+
+target_include_directories(${PROJECT_NAME} PRIVATE
+    ${ASCEND_PATH}/opp/built-in/op_proto/inc
+    ${INCLUDE_DIR}/graph
+    ${INCLUDE_DIR}/ge
+    ${INCLUDE_DIR}/register
+    ${INCLUDE_DIR}
+    ${INCLUDE_DIR}/external
+    ${CMAKE_CURRENT_BINARY_DIR}/generated_ops
+)
+
+target_link_directories(${PROJECT_NAME} PRIVATE
+    ${ASCEND_PATH}/compiler/lib64/stub
+    ${ASCEND_PATH}/compiler/lib64/ #todo delete  eager_style_graph_builder_base change to stub
+)
+target_link_options(${PROJECT_NAME} PRIVATE
+)
+target_link_libraries(${PROJECT_NAME} PRIVATE
+    -Wl,--verbose
+    -Wl,--no-as-needed
+    graph
+    register
+    ge_compiler
+    esb_generated_shared
+    eager_style_graph_builder_base
+    -Wl,--as-needed
+)
+
+add_dependencies(${PROJECT_NAME} esb_generated_shared)
\ No newline at end of file
diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/data/tf_forward.py b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/data/tf_forward.py
new file mode 100644
index 0000000000000000000000000000000000000000..f1227ad49811afff49c7abd7681dfaf3d62309e2
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/data/tf_forward.py
@@ -0,0 +1,63 @@
+"""
+# Copyright 2024 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+
+import tensorflow as tf
+import npu_bridge
+from tensorflow.core.protobuf.rewriter_config_pb2 import RewriterConfig
+import numpy as np
+
+def generate_tf_graph():
+    a = tf.compat.v1.placeholder(tf.float32, shape=[2, 3], name='a')
+    b = tf.compat.v1.placeholder(tf.float32, shape=[3, 2], name='b')
+    matmul = tf.linalg.matmul(a, b, name="matmul")
+    c = tf.compat.v1.placeholder(tf.float32, shape=[2, 2], name='c')
+    add = tf.add(matmul, c, name="add")
+    return tf.compat.v1.get_default_graph()
+
+def NetworkRun():
+    graph = generate_tf_graph()
+    input_a = graph.get_tensor_by_name('a:0')
+    input_b = graph.get_tensor_by_name('b:0')
+    input_c = graph.get_tensor_by_name('c:0')
+    output_nodes = graph.get_tensor_by_name('add:0')
+    a = np.array([[1.0, 2, 3], [4, 5, 6]])
+    b = np.array([[1.0, 2], [3, 4], [5, 6]])
+    c = np.array([[1.0, 1], [1, 1]])
+    
+    # 适配npu
+    config = tf.compat.v1.ConfigProto()
+    custom_op = config.graph_options.rewrite_options.custom_optimizers.add()
+
+    # 配置1：选择在昇腾AI处理器上执行推理
+    custom_op.name = "NpuOptimizer"
+    custom_op.parameter_map["use_off_line"].b = True
+
+    # 配置2：在线推理场景下建议保持默认值force_fp16，使用float16精度推理，以获得较优的性能
+    custom_op.parameter_map["precision_mode"].s = tf.compat.as_bytes("force_fp16")
+
+    # 配置3：图执行模式，推理场景下请配置为0，训练场景下为默认1
+    custom_op.parameter_map["graph_run_mode"].i = 0
+
+    # 配置4：关闭remapping和MemoryOptimizer
+    config.graph_options.rewrite_options.remapping = RewriterConfig.OFF
+    config.graph_options.rewrite_options.memory_optimization = RewriterConfig.OFF
+
+    with tf.compat.v1.Session(config=config, graph=graph) as sess:
+        out = sess.run(output_nodes, feed_dict={input_a:a, input_b:b, input_c:c})
+        print('---out---\n', out)
+
+if __name__=='__main__':
+    NetworkRun()
diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/CMakeLists.txt b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..60376b3fa427e1b72da7b3f64c77cc3afbafdb48
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/CMakeLists.txt
@@ -0,0 +1,74 @@
+message(STATUS "ASCEND_PATH: ${ASCEND_PATH}")
+
+###########  generate eager stype api  ######################
+set(CMAKE_CURRENT_BINARY_DIR ${PROJECT_SOURCE_DIR}/build)
+message(STATUS "Binary directory: ${CMAKE_CURRENT_BINARY_DIR}")
+set(common_compile_options
+        -std=c++17
+        -g
+        -Wall
+)
+
+set(common_compile_definitions
+        _GLIBCXX_USE_CXX11_ABI=0
+)
+# 1. 创建输出目录
+file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/generated_ops)
+
+# 2. 在构建目录创建占位源文件
+file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/generated_ops/placeholder.cpp
+        "// Placeholder file - will be replaced during build\n"
+        "void placeholder_function() {}\n")
+
+# 3. 使用 file(GLOB) 获取文件列表（但使用 CONFIGURE_DEPENDS 特性）
+file(GLOB GENERATED_CPP_FILES CONFIGURE_DEPENDS
+        "${CMAKE_CURRENT_BINARY_DIR}/generated_ops/*.cpp"
+)
+link_directories(${ASCEND_PATH}/libs)
+# 4. 定义代码生成命令
+add_custom_command(
+        OUTPUT
+        ${CMAKE_CURRENT_BINARY_DIR}/generated_ops/generated_code.flag
+        COMMAND ${CMAKE_COMMAND} -E remove -f ${CMAKE_CURRENT_BINARY_DIR}/generated_ops/*.cpp
+        COMMAND ${CMAKE_COMMAND} -E env ASCEND_OPP_PATH=${ASCEND_PATH}/ops env LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${ASCEND_PATH}/lib64 ${ASCEND_PATH}/bin/gen_esb --output_dir=${CMAKE_CURRENT_BINARY_DIR}/generated_ops
+        COMMAND ${CMAKE_COMMAND} -E touch ${CMAKE_CURRENT_BINARY_DIR}/generated_ops/generated_code.flag
+        COMMENT "Generating eager style graph builder code..."
+        VERBATIM
+)
+# 5. 创建自定义目标触发代码生成
+add_custom_target(generate_es_api ALL
+        DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/generated_ops/generated_code.flag)
+
+# 临时处理，后续run包发布了es api的so后，去掉这一步
+###########  generate eager stype api so  ######################
+
+add_library(esb_generated_shared SHARED
+        ${CMAKE_CURRENT_BINARY_DIR}/generated_ops/placeholder.cpp
+        ${GENERATED_CPP_FILES}
+)
+set_source_files_properties(${GENERATED_CPP_FILES}
+        PROPERTIES
+        GENERATED TRUE
+)
+add_dependencies(esb_generated_shared generate_es_api)
+target_compile_options(esb_generated_shared PRIVATE ${common_compile_options})
+target_compile_definitions(esb_generated_shared PRIVATE ${common_compile_definitions})
+
+target_include_directories(esb_generated_shared PUBLIC
+        ${CMAKE_CURRENT_BINARY_DIR}/generated_ops
+        ${PROJECT_SOURCE_DIR}/gen_es_api/temp
+        ${ASCEND_PATH}/opp/built-in/op_proto/inc
+        ${INCLUDE_DIR}/graph
+        ${INCLUDE_DIR}/ge
+        ${INCLUDE_DIR}/register
+        ${INCLUDE_DIR}
+        ${INCLUDE_DIR}/external
+)
+target_link_directories(esb_generated_shared PRIVATE
+        ${ASCEND_PATH}/compiler/lib64/stub
+        ${ASCEND_PATH}/compiler/lib64/
+)
+target_link_libraries(esb_generated_shared
+        PUBLIC
+        eager_style_graph_builder_base
+)
\ No newline at end of file
diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/compliant_node_builder.h b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/compliant_node_builder.h
new file mode 100644
index 0000000000000000000000000000000000000000..f08fda116d42a8ece38f2b422172009090478299
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/compliant_node_builder.h
@@ -0,0 +1,303 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_GRAPH_BUILDER_COMPLIANT_NODE_BUILDER_H_
+#define AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_GRAPH_BUILDER_COMPLIANT_NODE_BUILDER_H_
+#include <vector>
+#include <string>
+#include <utility>
+#include "graph/attr_value.h"
+#include "graph/gnode.h"
+#include "graph/operator.h"
+#include "ge_common/ge_api_types.h"
+
+namespace ge {
+/**
+ * @brief 从任意类型创建属性值的辅助函数
+ * @tparam T 输入类型
+ * @param t 输入值
+ * @return 创建的属性值对象
+ */
+template <typename T>
+AttrValue CreateFrom(T &&t) {
+  AttrValue av;
+  av.SetAttrValue(std::forward<T>(t));
+  return av;
+}
+
+/**
+ * @brief 合规节点构建器类，用于构建符合IR规范的图节点
+ *
+ * 该类提供了流式API来定义节点的IR输入、输出、属性和子图，
+ * 确保生成的节点符合图引擎的IR规范要求。
+ */
+class CompliantNodeBuilder {
+ public:
+  /**
+   * @brief 任意类型操作符类，继承自Operator
+   *
+   * 提供了对Operator类中IR相关方法的访问，包括动态输入/输出注册、
+   * 输入/输出注册、属性注册和子图注册等功能。
+   */
+  class AnyTypeOperator : public Operator {
+   public:
+    AnyTypeOperator(const char_t *name, const char_t *type) : Operator(name, type) {}
+    using Operator::DynamicInputRegister;
+    using Operator::InputRegister;
+    using Operator::OptionalInputRegister;
+
+    using Operator::DynamicOutputRegister;
+    using Operator::OutputRegister;
+
+    using Operator::AttrRegister;
+    using Operator::RequiredAttrWithTypeRegister;
+
+    using Operator::SubgraphRegister;
+  };
+
+  /**
+   * @brief IR属性类型枚举
+   */
+  enum IrAttrType {
+    kEsAttrRequired,  ///< 必需属性
+    kEsAttrOptional   ///< 可选属性
+  };
+
+  /**
+   * @brief 子图类型枚举
+   */
+  enum SubgraphType {
+    kEsSubgraphStatic,   ///< 静态子图
+    kEsSubgraphDynamic,  ///< 动态子图
+    kEsSubgraphEnd
+  };
+
+  /**
+   * @brief IR输入类型枚举
+   */
+  enum IrInputType {
+    kEsIrInputRequired,  ///< 必需输入
+    kEsIrInputOptional,  ///< 可选输入
+    kEsIrInputDynamic,   ///< 动态输入
+    kEsIrInputTypeEnd
+  };
+
+  /**
+   * @brief IR输出类型枚举
+   */
+  enum IrOutputType {
+    kEsIrOutputRequired,  ///< 必需输出
+    kEsIrOutputDynamic,   ///< 动态输出
+    kEsIrOutputTypeEnd
+  };
+
+  /**
+   * @brief IR属性定义结构体
+   */
+  struct IrAttrDef {
+    std::string attr_name;
+    IrAttrType ir_attr_type;
+    std::string attr_data_type;  // see `kIrAttrTypesMap` in `operator.cc`
+    AttrValue attr_default_value;
+  };
+
+  /**
+   * @brief IR输入定义结构体
+   */
+  struct IrInputDef {
+    std::string name;
+    IrInputType ir_input_type;
+    std::string symbol_id;
+  };
+
+  /**
+   * @brief IR输出定义结构体
+   */
+  struct IrOutputDef {
+    std::string name;
+    IrOutputType ir_output_type;
+    std::string symbol_id;
+  };
+
+  /**
+   * @brief IR子图定义结构体
+   */
+  struct IrSubgraphDef {
+    std::string name;
+    SubgraphType subgraph_type;
+  };
+
+ public:
+  /**
+   * @brief 构造函数
+   * @param graph 所属的图对象
+   */
+  explicit CompliantNodeBuilder(ge::Graph *graph) : owner_graph_(graph) {}
+
+  /**
+   * @brief 设置操作符类型
+   * @param type 操作符类型字符串
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &OpType(const char_t *type);
+
+  /**
+   * @brief 定义IR输入规范
+   * @param input_ir_def 输入IR定义向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &IrDefInputs(std::vector<IrInputDef> input_ir_def);
+
+  /**
+   * @brief 定义IR输出规范
+   * @param output_ir_def 输出IR定义向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &IrDefOutputs(std::vector<IrOutputDef> output_ir_def);
+
+  /**
+   * @brief 定义IR属性规范
+   * @param attr_ir_def 属性IR定义向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &IrDefAttrs(std::vector<IrAttrDef> attr_ir_def);
+
+  /**
+   * @brief 定义IR子图规范
+   * @param subgraph_ir_def 子图IR定义向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &IrDefSubgraphs(std::vector<IrSubgraphDef> subgraph_ir_def);
+
+  /**
+   * @brief 设置节点名称
+   * @param name 节点名称
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &Name(const char_t *name);
+
+  /**
+   * @brief 设置动态输入实例数量
+   * @param ir_name IR输入名称
+   * @param num 实例数量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceDynamicInputNum(const char_t *ir_name, int32_t num);
+
+  /**
+   * @brief 设置动态输出实例数量
+   * @param ir_name IR输出名称
+   * @param num 实例数量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceDynamicOutputNum(const char_t *ir_name, int32_t num);
+
+  /**
+   * @brief 设置输出数据类型
+   * @param name 输出名称
+   * @param data_type 数据类型
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceOutputDataType(const char_t *name, ge::DataType data_type);
+
+  /**
+   * @brief 设置输出形状
+   * @param name 输出名称
+   * @param shape 形状向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceOutputShape(const char_t *name, const std::vector<int64_t> &shape);
+
+  /**
+   * @brief 设置输出原始形状
+   * @param name 输出名称
+   * @param shape 原始形状向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceOutputOriginShape(const char_t *name, const std::vector<int64_t> &shape);
+
+  /**
+   * @brief 设置输出存储形状
+   * @param name 输出名称
+   * @param shape 存储形状向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceOutputStorageShape(const char_t *name, const std::vector<int64_t> &shape);
+
+  /**
+   * @brief 设置输出格式
+   * @param name 输出名称
+   * @param format 格式
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceOutputFormat(const char_t *name, ge::Format format);
+
+  /**
+   * @brief 设置输出原始格式
+   * @param name 输出名称
+   * @param format 原始格式
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceOutputOriginFormat(const char_t *name, ge::Format format);
+
+  /**
+   * @brief 设置输出存储格式
+   * @param name 输出名称
+   * @param format 存储格式
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceOutputStorageFormat(const char_t *name, ge::Format format);
+
+  /**
+   * @brief 设置子图实例
+   * @param name 子图名称
+   * @param sub_graphs 子图向量
+   * @return 当前构建器对象的引用，支持链式调用
+   */
+  CompliantNodeBuilder &InstanceSubgraphs(const char_t *name, const std::vector<std::string> &sub_graphs);
+
+  /**
+   * @brief 构建并返回图节点
+   * @return 构建完成的图节点对象
+   */
+  ge::GNode Build() const;
+
+ private:
+  void RegisterInputs(AnyTypeOperator &op) const;
+  void RegisterOutputs(AnyTypeOperator &op) const;
+  Status UpdateOutputDescs(AnyTypeOperator &op) const;
+  Status RegisterAttrs(AnyTypeOperator &op) const;
+  Status RegisterSubgraphs(AnyTypeOperator &op) const;
+
+ private:
+  ge::Graph *owner_graph_{nullptr};
+
+  // IR定义相关成员
+  std::string type_;
+  std::vector<IrInputDef> ir_def_inputs_;
+  std::vector<IrOutputDef> ir_def_outputs_;
+  std::vector<IrAttrDef> ir_def_attrs_;
+  std::vector<IrSubgraphDef> ir_def_subgraphs_;
+
+  // 实例信息相关成员
+  std::string name_;
+  std::unordered_map<std::string, int32_t> dynamic_input_ir_names_to_inst_num_;
+  std::unordered_map<std::string, int32_t> dynamic_output_ir_names_to_inst_num_;
+  std::unordered_map<std::string, TensorDesc> output_names_to_td_;
+  std::unordered_map<std::string, std::vector<std::string>> ir_names_to_subgraphs_;
+};
+}  // namespace ge
+
+#endif  // AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_GRAPH_BUILDER_COMPLIANT_NODE_BUILDER_H_
diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/es_c_graph_builder.h b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/es_c_graph_builder.h
new file mode 100644
index 0000000000000000000000000000000000000000..10f0df0fcae9bfa778257bda3f7ee761c5b68cd2
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/es_c_graph_builder.h
@@ -0,0 +1,120 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_GRAPH_BUILDER_ESC_GRAPH_BUILDER_H_
+#define AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_GRAPH_BUILDER_ESC_GRAPH_BUILDER_H_
+#include <list>
+#include <string>
+#include "graph/graph.h"
+#include "graph/c_types.h"
+#include "es_c_tensor_holder.h"
+struct EsCGraphBuilder {
+ public:
+  /**
+   * @brief 默认构造函数，创建一个名为"graph"的图构建器
+   */
+  EsCGraphBuilder() : EsCGraphBuilder("graph") {}
+
+  /**
+   * @brief 构造函数
+   * @param name 图名称
+   */
+  explicit EsCGraphBuilder(const char *name);
+
+  /**
+   * @brief 从图节点获取张量持有者
+   * @param node 图节点
+   * @param output_index 输出索引
+   * @return 张量持有者指针
+   */
+  EsCTensorHolder *GetTensorHolderFromNode(const ge::GNode &node, int32_t output_index);
+
+  /**
+   * @brief 在图的末尾添加图输入
+   * @param name 输入名称，可选
+   * @param type 输入类型，可选
+   * @return 张量持有者指针
+   */
+  EsCTensorHolder *AppendGraphInput(const ge::char_t *name = nullptr, const ge::char_t *type = nullptr);
+
+  /**
+   * @brief 在指定位置添加图输入
+   * @param index 输入索引位置
+   * @param name 输入名称，可选
+   * @param type 输入类型，可选
+   * @param data_type 数据类型，默认为C_DT_FLOAT
+   * @param format 数据格式，默认为C_FORMAT_ND
+   * @param dims 维度数组，可选
+   * @param dim_num 维度数量，默认为0
+   * @return 张量持有者指针
+   */
+  EsCTensorHolder *AddGraphInput(int32_t index, const ge::char_t *name = nullptr, const ge::char_t *type = nullptr,
+                                 C_DataType data_type = C_DT_FLOAT, C_Format format = C_FORMAT_ND,
+                                 const int64_t *dims = nullptr, int64_t dim_num = 0);
+
+  /**
+   * @brief 设置图输出
+   * @param tensor 张量持有者
+   * @param output_index 输出索引
+   * @return 操作状态
+   */
+  ge::Status SetGraphOutput(EsCTensorHolder *tensor, int32_t output_index);
+
+  /**
+   * @brief 获取内部的图对象
+   * @return 图对象指针
+   */
+  ge::Graph *GetGraph();
+
+  /**
+   * @brief 构建并返回图对象
+   * @return 构建完成的图对象
+   */
+  std::unique_ptr<ge::Graph> BuildGraph();
+
+  /**
+   * @brief 生成节点名称
+   * @param node_type 节点类型
+   * @return 生成的节点名称字符串
+   */
+  ge::AscendString GenerateNodeName(const ge::char_t *node_type);
+
+ private:
+  /**
+   * @brief 检查图是否有效
+   * @return 图是否有效
+   */
+  bool IsGraphValid() const;
+
+  /**
+   * @brief 从图节点获取张量持有者的内部实现
+   * @param node 图节点
+   * @param output_index 输出索引
+   * @return 张量持有者指针
+   */
+  EsCTensorHolder *GetTensorHolderFromNodeInner(const ge::GNode &node, int32_t output_index);
+
+ private:
+  int64_t NextNodeIndex();
+
+ private:
+  std::unique_ptr<ge::Graph> graph_;
+  std::list<std::unique_ptr<EsCTensorHolder>> tensors_holder_;
+  std::set<int32_t> graph_input_indexes_;
+  std::map<int32_t, EsCTensorHolder *> output_indexes_to_tensor_;
+  int64_t nodes_num_;
+};
+
+#endif  // AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_GRAPH_BUILDER_ESC_GRAPH_BUILDER_H_
diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/es_c_tensor_holder.h b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/es_c_tensor_holder.h
new file mode 100644
index 0000000000000000000000000000000000000000..db624bea2e2d54710458067f368f2902e9e5f262
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/gen_es_api/temp/es_c_tensor_holder.h
@@ -0,0 +1,121 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_EAGER_STYLE_GRAPH_BUILDER_ESC_TENSOR_HOLDER_H_
+#define AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_EAGER_STYLE_GRAPH_BUILDER_ESC_TENSOR_HOLDER_H_
+#include "graph/gnode.h"
+#include "ge_common/ge_api_types.h"
+#include "ge_common/ge_api_error_codes.h"
+struct EsCGraphBuilder;
+struct EsCTensorHolder {
+ public:
+  /**
+   * @brief 构造函数
+   * @param owner 所属的图构建器
+   * @param producer 生产者节点
+   * @param index 输出索引
+   *
+   * 注意：调用者需要保证传入的`producer`不为空，`index`合法
+   */
+  EsCTensorHolder(EsCGraphBuilder &owner, const ge::GNode &producer, int32_t index)
+      : owner_graph_builder_(owner), producer_(producer), producer_out_index_(index) {}
+
+  /**
+   * @brief 设置数据类型
+   * @param data_type 数据类型
+   * @return 操作状态
+   */
+  ge::Status SetDataType(const ge::DataType data_type);
+
+  /**
+   * @brief 设置数据格式
+   * @param format 数据格式
+   * @return 操作状态
+   */
+  ge::Status SetFormat(const ge::Format format);
+
+  /**
+   * @brief 设置原始数据格式
+   * @param format 原始数据格式
+   * @return 操作状态
+   */
+  ge::Status SetOriginFormat(const ge::Format format);
+
+  /**
+   * @brief 设置存储数据格式
+   * @param format 存储数据格式
+   * @return 操作状态
+   */
+  ge::Status SetStorageFormat(const ge::Format format);
+
+  /**
+   * @brief 设置原始形状
+   * @param shape 原始形状
+   * @return 操作状态
+   */
+  ge::Status SetOriginShape(const ge::Shape &shape);
+
+  /**
+   * @brief 设置存储形状
+   * @param shape 存储形状
+   * @return 操作状态
+   */
+  ge::Status SetStorageShape(const ge::Shape &shape);
+
+  /**
+   * @brief 设置形状（同时设置原始形状和存储形状）
+   * @param shape 形状
+   * @return 操作状态
+   */
+  ge::Status SetShape(const ge::Shape &shape);
+
+  /**
+   * @brief 设置原始符号形状
+   * @param shape_str 形状字符串数组
+   * @param shape_str_num 形状字符串数量
+   * @return 操作状态
+   */
+  ge::Status SetOriginSymbolShape(const char *const *shape_str, const int64_t shape_str_num);
+
+  /**
+   * @brief 获取生产者节点
+   * @return 生产者节点引用
+   */
+  ge::GNode &GetProducer();
+
+  /**
+   * @brief 获取输出索引
+   * @return 输出索引
+   */
+  int32_t GetOutIndex() const;
+
+  /**
+   * @brief 获取所属的图构建器
+   * @return 图构建器引用
+   */
+  EsCGraphBuilder &GetOwnerBuilder();
+
+ private:
+  ge::TensorDesc GetTd();
+
+  void UpdateTd(const ge::TensorDesc &td);
+
+  EsCGraphBuilder &owner_graph_builder_;
+  ge::GNode producer_;
+  int32_t producer_out_index_;
+};
+
+#endif  // AIR_CXX_COMPILER_GRAPH_EAGER_STYLE_EAGER_STYLE_GRAPH_BUILDER_ESC_TENSOR_HOLDER_H_
diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/readme.md b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/readme.md
new file mode 100644
index 0000000000000000000000000000000000000000..05016ec67ea14d50a24c6a65245b74557182a845
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/readme.md
@@ -0,0 +1,119 @@
+# 样例使用指导<a name="ZH-CN_TOPIC_0345664697"></a>
+
+## 功能描述<a name="section5991635456363"></a>
+
+本样例为MatMul+Add融合为GEMM自定义pass样例，使用TF在线推理演示框架如何调用自定义pass完成图优化。
+本样例使用eager style api和融合接口实现
+
+## 目录结构<a name="section7668345634665"></a>
+
+```
+├── src
+│   ├──fuse_matmul_add_pass.cpp  // pass实现文件 
+├── CMakeLists.txt               // 编译脚本
+├── data         
+|   ├──tf_forward.py             // TF在线构出原图后进行自定义pass和其他框架内置pass优化，然后执行优化后的图得到结果
+|—— gen_es_api
+|   |——temp                      // 临时目录，待头文件发布后删除
+|   |    |——compliant_node_builder.h
+|   |    |——es_c_graph_builder.h
+|   |    |——es_c_tensor_holder.h
+|   |——CMakeLists.txt            // 生成eager style api的编译脚本
+```
+
+## 环境要求<a name="section383335652346"></a>
+
+-   操作系统及架构：CentOS x86系统、CentOS aarch64系统、Euleros x86系统、Euleros aarch64系统
+-   编译器：gcc7及以上
+-   芯片：all
+-   python及依赖的库：python3.7.5、tensorflow1.15.0
+-   已完成昇腾AI软件栈在开发环境上的部署
+
+
+## 程序编译<a name="section6645633456813"></a>
+
+1. 根据实际情况修改**CMakeLists.txt**文件中的如下信息。
+
+   - ASCEND_PATH：指定到ATC或FwkACLlib的安装目录，例如/home/HwHiAiUser/Ascend/ascend-toolkit/latest
+
+   - target_include_directories：需要包含的头文件，对于本示例，无需修改。如果是用户自行开发的代码，当需要添加头文件时，在示例下方直接增加行即可，注意不要删除原有项目。如果网络中有自定义算子，请增加自定义算子的原型定义头文件。
+
+   - target_link_libraries：需要链接的库，对于本示例，无需修改。如果是用户自行开发的代码，当需要添加链接库时，在示例下方直接增加行即可，注意不要删除原有项目。
+
+     >禁止链接软件包中的其他so，否则后续升级可能会导致兼容性问题。
+
+2. 执行如下命令 生成eager style api
+
+   依次执行:
+
+   ```
+   mkdir build && cd build
+   cmake ..
+   make generate_es_api
+   ```
+   执行后，在**build**目录下产生generated_ops目录，内含es构图api的头文件及源码
+3. 执行如下命令生成es api对应的so，并拷贝到run包安装路径下
+   注意：此为临时步骤，待run包中发布了es api so，该步骤可删除
+   ```
+   make esb_generated_shared
+   ```
+   编译完成后，在**build**/gen_es_api目录下生成libesb_generated_shared.so 
+   将so拷贝到ASCEND_PATH下lib64目录下
+   ```
+   cp gen_es_api/libesb_generated_shared.so ${ASCEND_PATH}/x86_64-linux/lib64
+   ln -sf ${ASCEND_PATH}/x86_64-linux/lib64/libesb_generated_shared.so ${ASCEND_PATH}/latest/compiler/lib64/libesb_generated_shared.so
+   ```
+   若为arm环境，目标路径调整为
+   ```
+   cp gen_es_api/libesb_generated_shared.so ${ASCEND_PATH}/aarch64-linux/lib64
+   ln -sf ${ASCEND_PATH}/aarch64-linux/lib64/libesb_generated_shared.so ${ASCEND_PATH}/latest/compiler/lib64/libesb_generated_shared.so
+   ```
+4. 执行如下命令编译自定义pass so
+   ```
+   make
+   ```
+   编译结束后，在**build**目录下生成动态库文件**libfuse_matmul_add_pass.so**。
+   将**libfuse_matmul_add_pass.so**拷贝到$\{ASCEND\_PATH\}/opp/vendors/xxx/custom\_fusion\_passes/目录下。其中“xxx”为用户自定义目录。
+
+## 程序运行<a name="section4524573456563512"></a>
+
+1. 配置环境变量。
+
+   - 若运行环境上安装的“Ascend-cann-toolkit”包，环境变量设置如下：
+
+     ```
+     . ${HOME}/Ascend/ascend-toolkit/set_env.sh
+     ```
+
+     “$HOME/Ascend”请替换相关软件包的实际安装路径。
+
+2. 使用TF在线推理。
+
+   - 在线推理分别在目标文件夹下存放和不存放自定义pass so，执行如下命令：
+
+     **python3.7.5 tf_forward.py**
+
+     两次运行结果相同，结果展示：
+
+     ```
+     ---out---
+      [[23. 29.]
+      [50. 65.]]
+     ```
+
+   - 检查执行结果：
+
+     - 自定义pass生效前后运行结果相同。
+
+     - 自定义pass生效时对比npu编译过程中间dump图**ge_onnx_xxxxxxxx_graph_1_RunCustomPassBegin.pbtxt**和**ge_onnx_xxxxxxxx_graph_1_RunCustomPassEnd.pbtxt**发现模型已按照预期被优化。
+
+     - 日志中出现如下打印：
+
+       ```
+       EagerFuseMatMulAndAddPass begin.
+       Found MatMulAndAdd in graph.
+       Replace MatMulAndAdd to GEMM SUCCESS.
+       EagerFuseMatMulAndAddPass end.
+       ```
+
+       
\ No newline at end of file
diff --git a/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/src/fuse_matmul_add_pass.cpp b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/src/fuse_matmul_add_pass.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d9e0687da595e2907e87674ddffb97197595da84
--- /dev/null
+++ b/cplusplus/level1_single_api/3_ir/2_fuse_matmul_add_pass_in_eager_style/src/fuse_matmul_add_pass.cpp
@@ -0,0 +1,66 @@
+/**
+ * Copyright (c) Huawei Technologies Co., Ltd. 2025. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <iostream>
+#include "register_custom_pass.h"
+#include "es_all_ops.h"
+#include "pattern_matcher.h"
+#include "graph_rewriter.h"
+#include "pattern_fusion_pass.h"
+#include "fusion_pass_reg.h"
+
+using namespace ge;
+using namespace fusion;
+// |o>-----------------------------------
+// |o>    a  b
+// |o>    \ /              a   b    c
+// |o>   MatMul  c   ==>   \   |   /
+// |o>     \    /            GEMM
+// |o>      Add
+// |o>-----------------------------------
+// 融合说明：本例识别上图中左边的MatMul+Add结构并通过图修改接口替换为右边的单个GEMM节点
+// 改图接口返回值说明：本文件中的改图接口需要判断返回值, 基于可读性考虑除了pass入口函数外其他函数中的改图接口只接收返回值
+// 但不增加返回值处理代码。如需判断返回值，可配合使用custom_context.SetErrorMessage("xxx")方法
+class FuseMatMulAndAddInEagerPass : public PatternFusionPass {
+protected:
+    std::vector<std::unique_ptr<Graph>> PatternGraphs() override {
+        std::vector<std::unique_ptr<Graph>> pattern_graphs;
+        // build pattern graph
+        auto graph_builder = es::EsGraphBuilder("pattern");
+        auto a = graph_builder.CreateInput(0);
+        auto b = graph_builder.CreateInput(1);
+        auto c = graph_builder.CreateInput(2);
+
+        auto matmul = es::MatMulV2(a, b, nullptr, nullptr, false, false, false);
+        auto add = es::Add(c, matmul);
+        pattern_graphs.emplace_back(graph_builder.Build({add}));
+        return pattern_graphs;
+    }
+    bool MeetRequirements(const std::unique_ptr<MatchResult> &match_result) override {
+        return true;
+    }
+    std::unique_ptr<Graph> Replacement(const std::unique_ptr<MatchResult> &match_result) override {
+        auto replace_graph_builder = es::EsGraphBuilder("replacement");
+        auto r_a = replace_graph_builder.CreateInput(0);
+        auto r_b = replace_graph_builder.CreateInput(1);
+        auto r_c = replace_graph_builder.CreateInput(2);
+        auto alpha_const = replace_graph_builder.CreateScalar(1);
+        auto beta_const = replace_graph_builder.CreateScalar(1);
+        auto gemm = es::GEMM(r_a, r_b, r_c, alpha_const, beta_const);
+        return replace_graph_builder.Build({gemm});
+    }
+};
+REG_FUSION_PASS(FuseMatMulAndAddInEagerPass).Stage(CustomPassStage::kBeforeInferShape);
\ No newline at end of file