Skip to content

Applying Annotations

This document is organized into five sections describing how to work with various types of annotations in Binary Ninja using the API. Note that its companion documentation describes how to interact with many of the same elements via the UI.

  1. Symbols covers how to work with Symbols in a binary
  2. Types documents creating and interacting with types through the API
  3. Tags describes how to create tags and bookmarks
  4. Type Libraries explains how to work with Type Libraries, including multiple sources of information from which Binary Ninja can automatically source for type information from and how you can add to them
  5. Signature Libraries explains how to work with the signature library which match statically compiled functions which are then matched with type libraries


From an API perspective there are several helper functions available for working with Symbols. For example, to rename a function:

>>> = "newName"

Other objects or variables may need a symbol created and applied (this can also be done even if not using a helper function on objects such as functions):

>>> mysym = Symbol(SymbolType.FunctionSymbol, here, "myVariableName")
>>> mysym
<SymbolType.FunctionSymbol: "myVariableName" @ 0x80498d0>
>>> bv.define_user_symbol(mysym)

Note that here and bv are used in many of the previous examples. These shortcuts and several others are only available when running in the Binary Ninja python console and are used here for convenience.

Valid symbol types include:

SymbolType Description
FunctionSymbol Symbol for function that exists in the current binary
ImportAddressSymbol Symbol defined in the Import Address Table
ImportedFunctionSymbol Symbol for a function that is not defined in the current binary
DataSymbol Symbol for data in the current binary
ImportedDataSymbol Symbol for data that is not defined in the current binary
ExternalSymbol Symbols for data and code that reside outside the BinaryView
LibraryFunctionSymbol Symbols for functions identified as belonging to a shared library
SymbolicFunctionSymbol Symbols for functions without a concrete implementation or which have been abstractly represented
LocalLabelSymbol Symbol for a local label in the current binary


The tags API has been reworked in 3.4 to be easier to use and understand.


A TagType is the object that represents the "type" of tag something is. By default, a binary view starts off with several tag types, include "Bugs," "Crashes," "Important," and so on.

Before you can create a tag of a type that's not already in the BinaryView, you need to create the tag type.

Previously you'd have to handle this object and pass it around to other functions, but now you only need to remember its identifier:

>>> bv.create_tag_type("Really good looking code", "🦀")
<tag type Really good looking code: 🦀 >

Where now you won't need the TagType object to create new tags.

Creating Tags

There are three main types of tags: data tags, function tags, and address tags. All three of these use a similar .add_tag API.

Data Tags

Data tags are tags you create on a BinaryView, with a given address and some associated data. It also has optional fields for if you'd like to create this as a auto tag.

These tags can be at any address. If they happen to overlap with the address in a function, it will add the tag to the appropriate lines of disassembly/IL/decompilation. This is not the best way to get tags into a function, however.

>>> bv.add_tag(0xdeadbeef, "Vulnerabilities", "Some associated comment or serialized metadata")

Function and Address Tags

Function tags are tags you create on a Function with some associated data. Function tags appear at the top of a function and are a good way to label an entire function with some information.

>>> bv.add_tag_type("Cheese", "🧀")
<tag type Cheese: 🧀 >
>>> current_function.add_tag("Cheese", "This function smells funny")

If you include an address when you call Function.add_tag, you'll create an address tag. These are good for labeling specific instructions.

>>> current_function.add_tag("Bug", "This is comparing an unsigned number with a signed number!", 0xdeadbeef)

Auto Tags

Auto Tags are tags that are created automatically by some process or plugin. They will not be saved into the database as the assumption is that they are automatically created each time the file is to be opened. This is unlikely to be true for most use cases which is why the auto API was deprecated in favor of a unified API that defaults to auto=false.

If you want to create auto tags as of 3.4, provide the auto=True flag to the above calls.

>>> bv.add_tag(0xdeadbeef, "Vulnerability", "Buffer overflooooowwwww!!!", auto=True)


Binary Ninja provides a flexible API for creating and defining types explicitly. Binary Ninja Type objects are immutable, we provide two different methods for creating them. For simple types there are one-shot APIs that allow you to define the type with a single function call. Some situations are more complex and incremental construction is preferred, so we provide an additional TypeBuilder interface. The TypeBuilder interface is also useful for when you want to modify an existing type.

Simple Type Creation

There are a number of different type objects available for creation:

  • Integer Types
  • Characters Types (technically an integer)
  • Wide Characters Types (also technically an integer)
  • Boolean (guess what? also technically an integer)
  • Float Types (definitely not an integer)
  • Pointers
  • Void (like an integer but if its size was zero)
  • Functions
  • Arrays
  • Enumeration (kind of an integer)
  • Structures (probably has integers in it)
  • Type Definitions

Creating Types Using the Type Parser

Binary Ninja's built-in type parser provides a very convenient way of creating types when you can't remember the exact API to call. The parse_type_string API returns a tuple of Type and string name for the string passed.

>>> bv.parse_type_string("uint64_t")
(<type: uint64_t>, '')

Though convenient it is many orders of magnitude slower than simply calling the APIs directly. For applications where performance is at all desired the following APIs should be used.

Integer Types # Creates a 4 byte signed integer, False) # Creates an 8 bytes unsigned integer, altName="short") # Creates a 2 byte signed integer named 'short'
# Similarly through their classes directly
IntegerType.create(8, False)
IntegerType.create(2, altName="short")

Character Types

Type.char() # This is just a 1 byte signed integer and can be used as such
Type.wideChar(2, altName="wchar_t") # Creates a wide character with the name 'wchar_t'
# Similarly through their classes directly


# Similarly through its class directly

Floating-point Types

All floating point numbers are assumed to be signed

Type.float(4) # Creates a 4 byte ieee754 'float'
Type.float(8) # Creates a 8 byte ieee754 'double'
Type.float(10) # Creates a 10 byte ieee754 'long double'
Type.float(16) # Creates a 16 byte ieee754 'float128'
# Similarly through their classes directly


Type.void() # Create a void type which has zero size
# Similarly through its class directly


Type.pointer(bv.arch, # Create a pointer to a signed 4 byte integer
Type.pointer(, width=bv.arch.address_size) # Equivalent to the above but doesn't require an Architecture object be passed around
Type.pointer(bv.arch, Type.void(), const=True, volatile=False) # Creates a constant non volatile void pointer.
# Similarly through their classes directly
PointerType.create(, width=bv.arch.address_size)
PointerType.create(bv.arch, Type.void(), const=True, volatile=False)


Type.array(, 2) # Create an array of 2 - 4 byte integers
# Similarly through their classes directly
ArrayType.create(, 2)

Function Types

Type.function() # Creates a function with which takes no parameters and returns void
Type.function(Type.void(), [('arg1',]) # Create a function type which takes an integer as parameter and returns void
Type.function(params=[('arg1',]) # Same as the above
# Similarly through their classes directly
FunctionType.create(Type.void(), [('arg1',])

Create Anonymous Structures/Class/Unions

In Binary Ninja's type system supports anonymous and named structures. Anonymous structures are the simplest to understand, and what most people would expect.

Type.structure(members=[(, 'field_0')], type=StructureVariant.UnionStructureType) # Create a union with one integer members
Type.structure(members=[(, 'field_0'), (, 'field_4')], packed=True) # Created a packed structure containing two integer members
Type.structure(members=[(, 'field_0')], type=StructureVariant.ClassStructureType) # Create a class with one integer members
# Similarly through their classes directly
StructureType.create(members=[(, 'field_0')], type=StructureVariant.UnionStructureType)
StructureType.create(members=[(, 'field_0'), (, 'field_4')], packed=True)
StructureType.create(members=[(, 'field_0')], type=StructureVariant.ClassStructureType)

Create Anonymous Enumerations

Type.enumeration(members=[('ENUM_2', 2), ('ENUM_4', 4), ('ENUM_8', 8)])
Type.enumeration(members=['ENUM_0', 'ENUM_1', 'ENUM_2'])

Accessing Types

You may end up accessing types via a variety of APIs. In some cases, you're already working with a variable, function, or other object that has a type property:

>>> current_function.type
<type: immutable:FunctionTypeClass 'int32_t(int32_t argc, char** argv, char** envp)'>
>>> current_function.parameter_vars[2]
<var char** envp>
>>> current_function.parameter_vars[2].type
<type: immutable:PointerTypeClass 'char**'>

Explicit Type Lookup

Of course, there are also methods to directly access a type independent of its association with any given object in analysis:

>>> bv.types['Elf64_Header']
<type: immutable:StructureTypeClass 'struct Elf64_Header'>
# Or
>>> s = bv.get_type_by_name('Elf64_Header')
>>> s
<type: immutable:StructureTypeClass 'struct Elf64_Header'>
>>> s.members
[<struct Elf64_Ident ident, offset 0x0>, <enum e_type type, offset 0x10>, <enum e_machine machine, offset 0x12>, <uint32_t version, offset 0x14>, <void (* entry)(), offset 0x18>, <uint64_t program_header_offset, offset 0x20>, <uint64_t section_header_offset, offset 0x28>, <uint32_t flags, offset 0x30>, <uint16_t header_size, offset 0x34>, <uint16_t program_header_size, offset 0x36>, <uint16_t program_header_count, offset 0x38>, <uint16_t section_header_size, offset 0x3a>, <uint16_t section_header_count, offset 0x3c>, <uint16_t string_table, offset 0x3e>]

Accessing Data Variable Values

Even more powerful are the APIs when a type is applied to a specific data variable because you can directly query members or values according to the type:

>>> header = bv.get_data_var_at(bv.start)
>>> header
<var 0x0: struct Elf64_Header>
>>> header['ident']
<TypedDataAccessor type:struct Elf64_Ident value:{'signature': b'\x7fELF', 'file_class': 2, 'encoding': 1, 'version': 1, 'os': 0, 'abi_version': 0, 'pad': b'\x00\x00\x00\x00\x00\x00\x00'}>
>>> header['ident']['signature'].value

Important Concepts

Here's a few useful concepts when working Binary Ninja's type system.

Named Types

In Binary Ninja the name of a class/struct/union or enumeration is separate from its type definition. This is much like how it's done in C. The mapping between a structure's definition and its name is kept in the Binary View. Thus if we want to associate a name with our type we need an extra step.

bv.define_user_type('Foo', Type.structure(members=[(, 'field_0')]))

To reference a named type a NamedTypeReference object must be used. Say we want to add the struct Foo to our a new structure Bar.

t = Type.structure(members=[(, 'inner')])
n = 'Foo'
bv.define_user_type(n, t)
ntr = Type.named_type_from_registered_type(bv, n)
bv.define_user_type('Bar', Type.structure(members=[(ntr, 'outer')]))

The above is equivalent to the following C

struct Foo {
    int32_t inner;

struct Bar {
    struct Foo outer;

It is also possible to add the type referenced by Foo directly as an anonymous structure and thus there would be no need for a NamedTypeReference

t = Type.structure(members=[(, 'inner')])
bv.define_user_type('Bas', Type.structure(members=[(t, 'outer')]))

Yielding the following C equivalent:

struct Bas
    struct { int32_t inner; } outer;

Mutable Types

As Type objects are immutable, the Binary Ninja API provides a pure python implementation of types to provide mutability, these all inherit from MutableType and keep the same names as their immutable counterparts minus the Type part. Thus Structure is the mutable version of StructureType and Enumeration is the mutable version of MutableType. Type objects can be converted to MutableType objects using the Type.mutable_copy API and, MutableType objects can be converted to Type objects through the MutableType.immutable_copy API. Generally speaking you shouldn't need the mutable type variants for anything except creation of structures and enumerations, mutable type variants are provided for convenience and consistency. Building and defining a new structure can be done in a few ways. The first way would be the two step process of creating the structure then defining it.

s = StructureBuilder.create(members=[(IntegerType.create(4), 'field_0')])
bv.define_user_type('Foo', s)

A second option for more complicated situations you can opt for incremental initialization of the type:

s = StructureBuilder.create()
s.packed = True
bv.define_user_type('Foo', s)

Finally you can use the built-in context manager which automatically registers the created type with the provided BinaryView (bv) and name(Foo). Additionally when creating TypeLibraries a Type can be passed instead of a BinaryView

with StructureBuilder.builder(bv, 'Foo') as s:
    s.packed = True

Type Modification

Sometimes it's desired to modify a type which has already been registered with a Binary View. The hard way would be as follows:

  1. Look up the type you want to modify
  2. Get a mutable version of that type
  3. Modify the type how you wish
  4. Register that type with the Binary View again
s = bv.types['Foo'].mutable_copy()
bv.define_user_type('Foo', s)

This is a bit easier by using the builder context manager as follows:

with Type.builder(bv, 'Foo') as s:

Applying Types

There are 3 categories of object which can have Type objects applied to them.

  • Functions
  • Variables (i.e. local variables)
  • DataVariables (i.e. global variables)

As of the 3.0 API its much easier to apply types to Variables and DataVariables

Applying a type to a Function

Change a functions type to a function which returns void and takes zero parameters:

current_function.type = Type.function(Type.void(), [])

Applying a type to a Variable

Change the parameter of a function's zeroth parameter to a pointer to a character:

current_function.parameter_vars[0].type = Type.pointer(bv.arch, Type.char())

Applying a type to a DataVariable

>>> bv.get_data_var_at(here)
<var 0x408104: void>
>>> bv.get_data_var_at(here).type
<type: immutable:VoidTypeClass 'void'>
>>> bv.get_data_var_at(here).type = Type.char()

Applying a type where non exists yet

In some instances you may need to first create a DataVariable before you can set the type at a given location:

>>> bv.get_data_var_at(here)
# Nothing there yet!
>>> bv.define_user_data_var(here, "char*")
<var 0x22d50787c: char*>

Note that most of the APIs that take a Type object also take a C-style type string as a convenience helper as demonstrated as a difference between the last two examples.


Importing a header goes through the same code path as parsing source directly. You will just have to read the file and specify the appropriate command-line arguments as an array. See user type guide for directions for choosing arguments.

>>> with open('C:\projects\stdafx.h', 'r') as f:
...     TypeParser.default.parse_types_from_source(, 'stdafx.h', Platform['windows-x86_64'], [], ['--target=x86_64-pc-windows-msvc', '-x', 'c', '-std', 'c99', r'-isystemC:\Program Files (x86)\Microsoft Visual Studio\2019\Professional\VC\Tools\MSVC\14.28.29333\include', r'-isystemC:\Program Files (x86)\Windows Kits\10\Include\10.0.19041.0\ucrt', r'-isystemC:\Program Files (x86)\Windows Kits\10\Include\10.0.19041.0\shared', r'-isystemC:\Program Files (x86)\Windows Kits\10\Include\10.0.19041.0\um'])
(<types: [...], variables: [...], functions: [...]>, [])

Exporting a header uses the TypePrinter.print_all_types api, and outputs a string.

>>> print(TypePrinter.default.print_all_types(bv.types.items(), bv))
// Types for /bin/ls
// This header file generated by Binary Ninja

#include <stdint.h>
#include <stddef.h>
#include <stdlib.h>
#include <stdbool.h>
#include <wchar.h>

#define __packed
#define __noreturn
#define __convention(name)

Type Libraries

Type Library documentation has outgrown this section and now lives in a separate file.

Signature Library

While many signatures are built-in and require no interaction to automatically match functions, you may wish to add or modify your own. First, install the SigKit plugin from the plugin manager.

Running the signature matcher

The signature matcher runs automatically by default once analysis completes. You can turn this off in Settings > Analysis > Autorun Function Signature Matcher (or, analysis.signatureMatcher.autorun in Settings).

You can also trigger the signature matcher to run from the menu Tools > Run Analysis Module > Signature Matcher.

Once the signature matcher runs, it will print a brief report to the console detailing how many functions it matched and will rename matched functions. For example:

1 functions matched total, 0 name-only matches, 0 thunks resolved, 33 functions skipped because they were too small

Generating signature libraries

To generate a signature library for the currently-open binary, use Tools > Signature Library > Generate Signature Library. This will generate signatures for all functions in the binary that have a name attached to them. Note that functions with automatically-chosen names such as sub_401000 will be skipped. Once it's generated, you'll be prompted where to save the resulting signature library.

For headless users, you can generate signature libraries by using the sigkit API (examples and documentation). For more detailed information, see our blog post describing signature generation.

If you are accessing the sigkit API through the Binary Ninja GUI and you've installed the sigkit plugin through the plugin manager, you will need to import sigkit under a different name:

import Vector35_sigkit as sigkit

Installing signature libraries

Binary Ninja loads signature libraries from 2 locations:


Always place your signature libraries in your user directory. The install path is wiped whenever Binary Ninja auto-updates. You can locate it with Open Plugin Folder in the command palette and navigate "up" a directory.

Inside the signatures folder, each platform has its own folder for its set of signatures. For example, windows-x86_64 and linux-ppc32 are two sample platforms. When the signature matcher runs, it uses the signature libraries that are relevant to the current binary's platform. (You can check the platform of any binary you have open in the UI using the console and typing bv.platform)

Manipulating signature libraries

You can edit signature libraries programmatically using the sigkit API. A very basic example shows how to load and save signature libraries. Note that Binary Ninja only supports signatures in the .sig format; the other formats are for debugging. The easiest way to load and save signature libraries in this format are the sigkit.load_signature_library() and sigkit.save_signature_library() functions.

To help debug and optimize your signature libraries in a Signature Explorer GUI by using Tools > Signature Library > Explore Signature Library. This GUI can be opened through the sigkit API using sigkit.signature_explorer() and sigkit.explore_signature_library().

For a text-based approach, you can also export your signature libraries to JSON using the Signature Explorer. Then, you can edit them in a text editor and convert them back to a .sig using the Signature Explorer afterwards. Of course, these conversions are also accessible through the API as the sigkit.sig_serialize_json module, which provides a pickle-like interface. Likewise, sigkit.sig_serialize_fb provides serialization for the standard .sig format.