diff --git a/README.md b/README.md
new file mode 100644
index 0000000..c47b389
--- /dev/null
+++ b/README.md
@@ -0,0 +1,41 @@
+# C++ Bare Metal Senders and Receivers
+
+[![Unit Tests](https://github.com/intel/cpp-baremetal-senders-and-receivers/actions/workflows/unit_tests.yml/badge.svg)](https://github.com/intel/cpp-baremetal-senders-and-receivers/actions/workflows/unit_tests.yml)
+
+## WARNING: Work in Progress!
+
+This library is under active development. It is a work in progress and has not
+yet been proved in production. Proceed at your own risk.
+
+## Abstract
+
+*C++ Bare Metal Senders and Receivers* is a C++ header-only library that
+partially implements [P2300](https://wg21.link/p2300), the proposed C++ standard
+design for asynchronous computation.
+
+This implementation is intended for use on bare metal microcontrollers and
+implements a subset of the standard design. P2300 does encourage implementations
+not to do any runtime allocation, and that is fine for bare metal. But some
+notable differences here from the standard proposal are:
+
+- This implementation does not deal with exceptions in any way. An error is just
+  an error value.
+- Coroutines are not considered at all. This implementation does not preclude
+  coroutines as senders, but in a bare metal environment, the lack of
+  non-allocation guarantees currently makes using coroutines an uncertain
+  proposition at best.
+- Generalized tag dispatch may be missing for some functions.
+
+In general, although the P2300 design allows mechanism-agnostic handling of
+concurrency, the main mechanism of concurrency targeted by this library is
+interrupts. A careful reading of
+\[[intro.multithread](https://eel.is/c++draft/intro.multithread)\] is
+recommended to understand how the C++ memory model defines well-formed
+concurrent execution.
+
+C++20 is required. The following compilers are supported:
+ 
+- clang 14 through 17
+- gcc 12 through 13
+
+See the [full documentation](https://intel.github.io/cpp-baremetal-senders-and-receivers/).