Apps

The user-facing surfaces of TensorFlow. "Apps" here means language frontends plus TensorFlow Lite — the things people actually import or link against. Internal runtime components are documented under systems, and compilers under compilers.

Page	Surface	Path
Python API	import tensorflow as tf	tensorflow/python/
C API	tensorflow/c/c_api.h (stable ABI)	tensorflow/c/
C++ API	tensorflow::Scope, op builders	tensorflow/cc/
Java / Android API	org.tensorflow.*	tensorflow/java/
Go API	github.com/tensorflow/tensorflow/tensorflow/go	tensorflow/go/
Keras (bundled v2)	tf.keras	tensorflow/python/keras/
TensorFlow Lite	Mobile/embedded runtime + delegates	tensorflow/lite/

Other tiny surfaces not given their own page:

• tensorflow/js/ — a small TFJS converter shim; the actual TFJS framework is in tensorflow/tfjs (separate repo).
• tensorflow/dtensor/ — DTensor public API; covered as a system in systems/distribution-strategy.
• tensorflow/examples/ — sample programs that exercise the APIs but aren't APIs themselves.

How frontends layer on the runtime

graph TD
    PyAPI[Python tf.* package]
    CcAPI[C++ tensorflow::ops]
    JavaAPI[org.tensorflow]
    GoAPI[Go tensorflow]

    CAPI[C API tensorflow/c/c_api.h]

    Eager[Eager runtime tensorflow/core/common_runtime/eager]
    Graph[Direct/master Session]

    PyAPI -- pybind11 / SWIG --> Eager
    PyAPI -- pybind11 / SWIG --> Graph
    PyAPI -. uses .-> CAPI
    CcAPI --> Graph
    CcAPI --> Eager
    JavaAPI --> CAPI
    GoAPI --> CAPI

The C API is the stable ABI: every other-language binding (Java, Go, Rust, Haskell, etc.) links against libtensorflow.so and uses c_api.h. Python is the exception — it talks directly to the runtime via pybind11 wrappers (tensorflow/python/_pywrap_tensorflow*.cc, tensorflow/python/tfe_wrapper.cc) for performance and feature coverage.

Versions of the API

• tf.* (v2) — eager-by-default, the modern entry point. Everything in this wiki targets v2.
• tf.compat.v1.* — the v1-style API still imported through a compatibility shim (tensorflow/compat_template.__init__.py, tensorflow/python/compat/). Used when you need Session, Graph-only mode, tf.placeholder, etc.
• tf.experimental.* — APIs subject to change. Living in tensorflow/python/ under various directories.

The PUBLIC API surface is determined by @tf_export(...) decorators and codified in tensorflow/tools/api/golden/ golden files.