diff --git a/README.md b/README.md
index 65c0659..2c5776e 100644
--- a/README.md
+++ b/README.md
@@ -13,6 +13,7 @@ Another serialization system: minimalist and canonicalization.
 - [Encoding Spec](docs/encoding_spec.md)
 - [Schema Language](docs/schema_language.md)
 - [Real-World Examples](docs/real_world_examples.md)
+- [API](docs/molecule_api.md)
 
 ## Features
 * `default` — Default features: `std`, utilizes `faster-hex` for hexadecimal operations and enables [bytes] standard features.
diff --git a/docs/molecule_api.md b/docs/molecule_api.md
new file mode 100644
index 0000000..acbb0fb
--- /dev/null
+++ b/docs/molecule_api.md
@@ -0,0 +1,85 @@
+## Molecule API
+
+Molecule is based on schema serialization. The serialization format generated by any language based on the same schema should be consistent. The most basic API for multiple languages ​​should include data verification, schema compatibility reading, serialization and deserialization, and obtaining raw data (excluding the molecule header).
+
+### Rust API
+
+In the basic implementation of rust, the generated code uses three traits to implement three different functions of a structure:
+
+#### Entity
+```rust
+pub trait Entity: fmt::Debug + Default + Clone {
+    type Builder: Builder;
+    const NAME: &'static str;
+    fn new_unchecked(data: Bytes) -> Self;
+    fn as_bytes(&self) -> Bytes;
+    fn as_slice(&self) -> &[u8];
+    fn from_slice(slice: &[u8]) -> VerificationResult<Self>;
+    fn from_compatible_slice(slice: &[u8]) -> VerificationResult<Self>;
+    fn new_builder() -> Self::Builder;
+    fn as_builder(self) -> Self::Builder;
+}
+```
+Entity corresponds to a serialized data structure. There are two APIs that are easily confused: `as_slice` defined in the trait and `raw_data` that comes with its own structure.
+
+`as_slice`: Complete molecule format data
+`raw_data`: Original data, i.e. without the molecule header
+
+Molecule supports compatible reading of data. The so-called compatibility refers to a structure like table, which can dynamically add fields. The old schema can read the data generated by the schema after adding fields, but it does not support deleting fields.
+
+```mol
+vector Inner <byte>;
+
+table Old {
+    a: Inner,
+    b: Inner,
+}
+
+table New {
+    a: Inner,
+    b: Inner,
+    c: Inner,
+}
+```
+
+Like the scheme in the example above, `Old` can use the `from_compatible_slice` api to read `New's` data.At the same time, the structure will have APIs such as `count_extra_fields`/`has_extra_fields`/`has_extra_fields` to let users know that there is extra data in the read data. It is currently compatible with reading.
+
+#### Reader
+```rust
+pub trait Reader<'r>: Sized + fmt::Debug + Clone + Copy {
+    type Entity: Entity;
+    const NAME: &'static str;
+    fn verify(slice: &[u8], compatible: bool) -> VerificationResult<()>;
+    fn new_unchecked(slice: &'r [u8]) -> Self;
+    fn as_slice(&self) -> &'r [u8];
+    fn from_slice(slice: &'r [u8]) -> VerificationResult<Self> {
+        Self::verify(slice, false).map(|_| Self::new_unchecked(slice))
+    }
+    fn from_compatible_slice(slice: &'r [u8]) -> VerificationResult<Self> {
+        Self::verify(slice, true).map(|_| Self::new_unchecked(slice))
+    }
+    fn to_entity(&self) -> Self::Entity;
+}
+```
+
+Each structure will generate at least one corresponding `Reader` structure, which has the ability to obtain the field data inside the structure.
+
+
+#### Builder
+
+```rust
+pub trait Builder: Default {
+    type Entity: Entity;
+    const NAME: &'static str;
+    fn expected_length(&self) -> usize;
+    fn write<W: io::Write>(&self, writer: &mut W) -> io::Result<()>;
+    fn build(&self) -> Self::Entity;
+}
+```
+
+Builder is the key to building and serializing the molecule structure. Use the builder mode to generate a builder structure, put all fields into it, and convert it into a serialized structure(Entity) through the `build` API.
+
+
+#### Union
+
+The union structure will have two more structures than other structures when generating code, corresponding to different internal data types, and ends with Union.
diff --git a/tools/codegen/src/generator/languages/rust/builder/implementation.rs b/tools/codegen/src/generator/languages/rust/builder/implementation.rs
index e396940..f7a138e 100644
--- a/tools/codegen/src/generator/languages/rust/builder/implementation.rs
+++ b/tools/codegen/src/generator/languages/rust/builder/implementation.rs
@@ -278,6 +278,12 @@ fn gen_from_iter(name: &str, item_name: &str) -> m4::TokenStream {
                     Self::new_builder().extend(iter.into_iter().map(Into::into)).build()
                 }
             }
+
+            impl From<Vec<u8>> for #entity {
+                fn from(v: Vec<u8>) -> Self {
+                    Self::new_builder().set(v.into_iter().map(Into::into).collect()).build()
+                }
+            }
         )
     } else {
         quote!()
@@ -290,6 +296,13 @@ fn gen_from_iter(name: &str, item_name: &str) -> m4::TokenStream {
             }
         }
 
+        impl From<Vec<#item_name>> for #entity
+        {
+            fn from(v: Vec<#item_name>) -> Self {
+                Self::new_builder().set(v).build()
+            }
+        }
+
         #maybe_byte_vec
     )
 }
diff --git a/tools/codegen/src/generator/languages/rust/builder/setters.rs b/tools/codegen/src/generator/languages/rust/builder/setters.rs
index 4d32e86..851a025 100644
--- a/tools/codegen/src/generator/languages/rust/builder/setters.rs
+++ b/tools/codegen/src/generator/languages/rust/builder/setters.rs
@@ -119,9 +119,7 @@ fn impl_setters_for_vector(inner_name: &str) -> m4::TokenStream {
             self
         }
         pub fn extend<T: ::core::iter::IntoIterator<Item=#inner>>(mut self, iter: T) -> Self {
-            for elem in iter {
-                self.0.push(elem);
-            }
+            self.0.extend(iter);
             self
         }
         pub fn replace(&mut self, index: usize, v: #inner) -> Option<#inner> {