diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..2c96eb1
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,2 @@
+target/
+Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
new file mode 100644
index 0000000..6650136
--- /dev/null
+++ b/Cargo.toml
@@ -0,0 +1,6 @@
+[workspace]
+
+members = [
+  "haskell-ffi",
+  "haskell-ffi-derive",
+]
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..0701c68
--- /dev/null
+++ b/README.md
@@ -0,0 +1,2 @@
+# Rust library for easy interop with Haskell
+
diff --git a/haskell-ffi-derive/Cargo.toml b/haskell-ffi-derive/Cargo.toml
new file mode 100644
index 0000000..1ca1dc5
--- /dev/null
+++ b/haskell-ffi-derive/Cargo.toml
@@ -0,0 +1,12 @@
+[package]
+name = "haskell-ffi-derive"
+version = "0.1.0"
+edition = "2021"
+
+[lib]
+proc-macro = true
+
+[dependencies]
+syn = "1.0"
+quote = "1.0"
+proc-macro2 = "1.0"
\ No newline at end of file
diff --git a/haskell-ffi-derive/src/lib.rs b/haskell-ffi-derive/src/lib.rs
new file mode 100644
index 0000000..671a594
--- /dev/null
+++ b/haskell-ffi-derive/src/lib.rs
@@ -0,0 +1,85 @@
+//! Macro for deriving `HaskellSize` instances for structs
+//!
+//! Implementation is adapted from the `heapsize` example in the `syn` crate.
+//! The implementation is not identical, however: `haskell_size` does not take
+//! any value as input, but is entirely type-based.
+
+use proc_macro2::TokenStream;
+use quote::quote;
+use syn::{
+    parse_macro_input, parse_quote, punctuated::Iter, Data, DeriveInput, Field, Fields,
+    GenericParam, Generics,
+};
+
+/// Derive `HaskellSize` instance
+///
+/// NOTE: Only structs are currently supported.
+#[proc_macro_derive(HaskellSize)]
+pub fn haskell_size_derive(input: proc_macro::TokenStream) -> proc_macro::TokenStream {
+    // Parse the input tokens into a syntax tree.
+    let input: DeriveInput = parse_macro_input!(input as DeriveInput);
+
+    // Used in the quasi-quotation below as `#name`.
+    let name = &input.ident;
+
+    // Add a bound `T: HaskellSize` to every type parameter T.
+    let without_tag: Generics = add_trait_bounds(input.generics);
+
+    // The instance itself must get an additional `Tag` argument
+    //
+    // NOTE: Things will go badly if one of the user's parameters is also named `Tag`.
+    let mut including_tag: Generics = without_tag.clone();
+    including_tag
+        .params
+        .push(GenericParam::Type(parse_quote!(Tag)));
+
+    let (including_tag_impl, _, _) = including_tag.split_for_impl();
+    let (_, without_tag_tys, without_tag_where) = without_tag.split_for_impl();
+
+    // Generate an expression to sum up the size of each field.
+    let sum = haskell_size_sum(&input.data);
+
+    let expanded = quote! {
+        impl #including_tag_impl HaskellSize<Tag> for #name #without_tag_tys #without_tag_where {
+            fn haskell_size(tag: PhantomData<Tag>) -> usize {
+                #sum
+            }
+        }
+    };
+
+    // Hand the output tokens back to the compiler.
+    proc_macro::TokenStream::from(expanded)
+}
+
+/// Add a bound `T: HaskellSize<Tag>` to every type parameter T.
+fn add_trait_bounds(mut generics: Generics) -> Generics {
+    for param in &mut generics.params {
+        if let GenericParam::Type(ref mut type_param) = *param {
+            type_param.bounds.push(parse_quote!(HaskellSize<Tag>));
+        }
+    }
+    generics
+}
+
+/// Generate an expression to sum up the size of each field.
+fn haskell_size_sum(data: &Data) -> TokenStream {
+    match data {
+        Data::Struct(ref data) => match &data.fields {
+            Fields::Named(fields) => haskell_size_fields(fields.named.iter()),
+            Fields::Unnamed(fields) => haskell_size_fields(fields.unnamed.iter()),
+            Fields::Unit => quote!(0),
+        },
+        Data::Enum(_) | Data::Union(_) => unimplemented!(),
+    }
+}
+
+/// Auxiliary to `haskell_size_sum`
+fn haskell_size_fields(fields: Iter<Field>) -> TokenStream {
+    let recurse = fields.map(|f| {
+        let t = &f.ty;
+        quote! { <#t> :: haskell_size(tag) }
+    });
+    quote! {
+        0 #(+ #recurse)*
+    }
+}
diff --git a/haskell-ffi/Cargo.toml b/haskell-ffi/Cargo.toml
new file mode 100644
index 0000000..aee7d20
--- /dev/null
+++ b/haskell-ffi/Cargo.toml
@@ -0,0 +1,13 @@
+[package]
+name = "haskell-ffi"
+version = "0.1.0"
+edition = "2021"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+bincode = "1.3"
+borsh = "0.9"
+haskell-ffi-derive = { path = "../haskell-ffi-derive" }
+ref-cast = "1.0"
+serde = "1.0"
diff --git a/haskell-ffi/src/bincode.rs b/haskell-ffi/src/bincode.rs
new file mode 100644
index 0000000..86ec825
--- /dev/null
+++ b/haskell-ffi/src/bincode.rs
@@ -0,0 +1,36 @@
+use std::{
+    io::{Error, ErrorKind, Write},
+    marker::PhantomData,
+};
+
+/// Implement `to_haskell` using `bincode`
+///
+/// The result will be length-prefixed ("bincode-in-Borsh").
+pub fn bincode_to_haskell<Tag, T, W>(
+    t: &T,
+    writer: &mut W,
+    _: PhantomData<Tag>,
+) -> Result<(), Error>
+where
+    T: serde::ser::Serialize,
+    W: Write,
+{
+    match bincode::serialize(t) {
+        Ok(vec) => borsh::BorshSerialize::serialize(&vec, writer),
+        Err(e) => Err(Error::new(ErrorKind::InvalidData, e)),
+    }
+}
+
+/// Implement `from_haskell` using `bincode`
+///
+/// See als `bincode_to_haskell`
+pub fn bincode_from_haskell<Tag, T>(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<T, Error>
+where
+    T: serde::de::DeserializeOwned,
+{
+    let vec: Vec<u8> = borsh::BorshDeserialize::deserialize(buf)?;
+    match bincode::deserialize(vec.as_ref()) {
+        Ok(x) => Ok(x),
+        Err(e) => Err(Error::new(ErrorKind::InvalidData, e)),
+    }
+}
diff --git a/haskell-ffi/src/deriving_via.rs b/haskell-ffi/src/deriving_via.rs
new file mode 100644
index 0000000..2182360
--- /dev/null
+++ b/haskell-ffi/src/deriving_via.rs
@@ -0,0 +1,113 @@
+use borsh::{BorshDeserialize, BorshSerialize};
+use ref_cast::RefCast;
+use std::{
+    cmp::Ordering,
+    fmt::Debug,
+    hash::{Hash, Hasher},
+    io::{Error, Write},
+    marker::PhantomData,
+};
+
+use crate::{from_haskell::FromHaskell, to_haskell::ToHaskell};
+
+/*******************************************************************************
+  Deriving-via support
+*******************************************************************************/
+
+#[derive(RefCast)]
+#[repr(transparent)]
+/// Newtype for "deriving-via" instances
+///
+/// The purpose of this newtype is best illustrated through its instances:
+///
+/// ```ignore
+/// impl<Tag, T: ToHaskell<Tag>>   BorshSerialize   for Haskell<Tag, T>
+/// impl<Tag, T: FromHaskell<Tag>> BorshDeserialize for Haskell<Tag, T>
+/// ```
+///
+/// This is primarily used internally: when deriving `ToHaskell`/`FromHaskell`
+/// instances for standard types, we want to re-use the logic from `borsh`,
+/// rather than re-implement everything here. We do this by turning say a
+/// `Vec<T>` into a `Vec<Haskell<Tag, T>>`, and then call functions from
+/// `borsh`. The use of the newtype wrapper then ensures that the constraint
+/// on `T` will be in terms of `ToHaskell`/`FromHaskell` again.
+pub struct Haskell<Tag, T>(pub T, PhantomData<Tag>);
+
+pub fn tag_val<Tag, T>(t: T) -> Haskell<Tag, T> {
+    Haskell(t, PhantomData)
+}
+
+pub fn tag_ref<Tag, T>(t: &T) -> &Haskell<Tag, T> {
+    RefCast::ref_cast(t)
+}
+
+pub fn untag_val<Tag, T>(tagged: Haskell<Tag, T>) -> T {
+    tagged.0
+}
+
+pub fn untag_ref<Tag, T>(tagged: &Haskell<Tag, T>) -> &T {
+    &tagged.0
+}
+
+/*******************************************************************************
+  Standard instances
+*******************************************************************************/
+
+impl<Tag, T: Debug> Debug for Haskell<Tag, T> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        self.0.fmt(f)
+    }
+}
+
+impl<Tag, T: PartialEq> PartialEq for Haskell<Tag, T> {
+    fn eq(&self, other: &Self) -> bool {
+        self.0 == other.0
+    }
+}
+
+impl<Tag, T: Eq> Eq for Haskell<Tag, T> {}
+
+impl<Tag, T: PartialOrd> PartialOrd for Haskell<Tag, T> {
+    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+        self.0.partial_cmp(&other.0)
+    }
+}
+
+impl<Tag, T: Hash> Hash for Haskell<Tag, T> {
+    fn hash<H: Hasher>(&self, state: &mut H) {
+        self.0.hash(state);
+    }
+}
+
+impl<Tag, T: Default> Default for Haskell<Tag, T> {
+    fn default() -> Self {
+        Self(Default::default(), PhantomData)
+    }
+}
+
+impl<Tag, T: Clone> Clone for Haskell<Tag, T> {
+    fn clone(&self) -> Self {
+        Self(self.0.clone(), PhantomData)
+    }
+}
+
+impl<Tag, T: Copy> Copy for Haskell<Tag, T> {}
+
+/*******************************************************************************
+  Forwarding instances
+
+  NOTE: We do not expect _additional_ forwarding instances to be defined.
+*******************************************************************************/
+
+impl<Tag, T: ToHaskell<Tag>> BorshSerialize for Haskell<Tag, T> {
+    fn serialize<W: Write>(&self, writer: &mut W) -> Result<(), Error> {
+        self.0.to_haskell(writer, PhantomData)
+    }
+}
+
+impl<Tag, T: FromHaskell<Tag>> BorshDeserialize for Haskell<Tag, T> {
+    fn deserialize(buf: &mut &[u8]) -> std::io::Result<Self> {
+        let tag: PhantomData<Tag> = PhantomData;
+        T::from_haskell(buf, tag).map(tag_val)
+    }
+}
diff --git a/haskell-ffi/src/from_haskell.rs b/haskell-ffi/src/from_haskell.rs
new file mode 100644
index 0000000..277e539
--- /dev/null
+++ b/haskell-ffi/src/from_haskell.rs
@@ -0,0 +1,76 @@
+use std::{
+    io::{Error, ErrorKind},
+    marker::PhantomData,
+};
+
+use crate::HaskellSize;
+
+/*******************************************************************************
+  Main class definition
+*******************************************************************************/
+
+const ERROR_NOT_ALL_BYTES_READ: &str = "Not all bytes read";
+
+pub trait FromHaskell<Tag>: Sized {
+    /// Deserialize data sent from Haskell
+    ///
+    /// This is the analogue of `BorshDeserialize::deserialize`.
+    //
+    /// See `ToHaskell` for a detailed discussion of the `tag` argument.
+    fn from_haskell(buf: &mut &[u8], tag: PhantomData<Tag>) -> Result<Self, Error>;
+
+    fn from_haskell_slice(slice: &[u8], tag: PhantomData<Tag>) -> Result<Self, Error> {
+        let mut slice_mut = slice;
+        let result = Self::from_haskell(&mut slice_mut, tag)?;
+        if !slice_mut.is_empty() {
+            return Err(Error::new(ErrorKind::InvalidData, ERROR_NOT_ALL_BYTES_READ));
+        }
+        Ok(result)
+    }
+}
+
+/*******************************************************************************
+  Derived functionality
+
+  See comments in `to_haskell` for why these functions do not live inside the
+  trait.
+*******************************************************************************/
+
+/// Marshall value with variable-sized encoding
+pub fn marshall_from_haskell_var<Tag, T>(inp: *const u8, len: usize, tag: PhantomData<Tag>) -> T
+where
+    T: FromHaskell<Tag>,
+{
+    let mut vec: Vec<u8> = vec![0; len];
+    unsafe {
+        std::ptr::copy(inp, vec.as_mut_ptr(), len);
+    }
+    match T::from_haskell_slice(vec.as_ref(), tag) {
+        Ok(t) => t,
+        Err(e) => panic!("{}", e),
+    }
+}
+
+/// Marshall value with fixed-size encoding
+///
+/// The `len` argument here is only to verify that the Haskell-side and
+/// Rust-side agree on the size of the encoding.
+pub fn marshall_from_haskell_fixed<Tag, T>(
+    inp: *const u8,
+    inp_len: usize,
+    tag: PhantomData<Tag>,
+) -> T
+where
+    T: FromHaskell<Tag> + HaskellSize<Tag>,
+{
+    let expected_len = T::haskell_size(tag);
+
+    if inp_len != expected_len {
+        panic!(
+            "expected buffer of size {}, but got {}",
+            expected_len, inp_len
+        )
+    } else {
+        marshall_from_haskell_var(inp, inp_len, tag)
+    }
+}
diff --git a/haskell-ffi/src/haskell_size.rs b/haskell-ffi/src/haskell_size.rs
new file mode 100644
index 0000000..810db74
--- /dev/null
+++ b/haskell-ffi/src/haskell_size.rs
@@ -0,0 +1,223 @@
+use std::marker::PhantomData;
+
+use crate::{derive_size_tuple_instance, fold_types};
+
+pub use haskell_ffi_derive::HaskellSize;
+
+/*******************************************************************************
+  Main class definition
+*******************************************************************************/
+
+pub trait HaskellSize<Tag> {
+    /// Statically known size (in bytes)
+    fn haskell_size(tag: PhantomData<Tag>) -> usize;
+}
+
+/*******************************************************************************
+  Simple instances
+
+  Note: the following types in the Borsh spec do _not_ have statically known sizes:
+
+  - Vec<T>
+  - HashMap<K, V>
+  - HashSet<T>
+  - Option<T>
+  - String
+*******************************************************************************/
+
+impl<Tag> HaskellSize<Tag> for u8 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        1
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for u16 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        2
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for u32 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        4
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for u64 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        8
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for u128 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        16
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for i8 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        1
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for i16 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        2
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for i32 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        4
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for i64 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        8
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for i128 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        16
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for f32 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        4
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for f64 {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        8
+    }
+}
+
+impl<Tag> HaskellSize<Tag> for () {
+    fn haskell_size(_tag: PhantomData<Tag>) -> usize {
+        0
+    }
+}
+
+impl<Tag, T: HaskellSize<Tag>, const N: usize> HaskellSize<Tag> for [T; N] {
+    fn haskell_size(tag: PhantomData<Tag>) -> usize {
+        T::haskell_size(tag) * N
+    }
+}
+
+/*******************************************************************************
+  Tuples
+
+  We support the same sizes of tuples as `borsh` does.
+*******************************************************************************/
+
+derive_size_tuple_instance!(T0, T1);
+derive_size_tuple_instance!(T0, T1, T2);
+derive_size_tuple_instance!(T0, T1, T2, T3);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14);
+derive_size_tuple_instance!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15);
+derive_size_tuple_instance!(
+    T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16
+);
+derive_size_tuple_instance!(
+    T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16, T17
+);
+derive_size_tuple_instance!(
+    T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16, T17, T18
+);
+derive_size_tuple_instance!(
+    T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16, T17, T18, T19
+);
+
+/*******************************************************************************
+  Sanity checks
+*******************************************************************************/
+
+#[cfg(test)]
+mod tests {
+    use std::io::Error;
+
+    use borsh::BorshSerialize;
+
+    use super::*;
+
+    enum ExampleTag {}
+
+    #[derive(HaskellSize, BorshSerialize)]
+    struct EmptyStruct;
+
+    #[derive(HaskellSize, BorshSerialize)]
+    struct UnnamedStruct(u16, (u8, u32));
+
+    #[derive(HaskellSize, BorshSerialize)]
+    struct NamedStruct {
+        a: u8,
+        b: u16,
+        c: (u32, u64),
+    }
+
+    #[derive(HaskellSize, BorshSerialize)]
+    struct ParamStruct<T> {
+        a: u8,
+        b: (T, T, T),
+    }
+
+    #[test]
+    fn empty() -> Result<(), Error> {
+        let tag: PhantomData<ExampleTag> = PhantomData;
+        assert_eq!(EmptyStruct::haskell_size(tag), 0);
+        let encoded = EmptyStruct.try_to_vec()?;
+        assert_eq!(encoded.len(), EmptyStruct::haskell_size(tag));
+        Ok(())
+    }
+
+    #[test]
+    fn unnamed() -> Result<(), Error> {
+        let tag: PhantomData<ExampleTag> = PhantomData;
+        assert_eq!(UnnamedStruct::haskell_size(tag), 7);
+        let encoded = UnnamedStruct(1, (2, 3)).try_to_vec()?;
+        assert_eq!(encoded.len(), UnnamedStruct::haskell_size(tag));
+        Ok(())
+    }
+
+    #[test]
+    fn named() -> Result<(), Error> {
+        let tag: PhantomData<ExampleTag> = PhantomData;
+        assert_eq!(NamedStruct::haskell_size(tag), 15);
+        let encoded = NamedStruct {
+            a: 1,
+            b: 2,
+            c: (3, 4),
+        }
+        .try_to_vec()?;
+        assert_eq!(encoded.len(), NamedStruct::haskell_size(tag));
+        Ok(())
+    }
+
+    #[test]
+    fn param() -> Result<(), Error> {
+        let tag: PhantomData<ExampleTag> = PhantomData;
+        assert_eq!(<ParamStruct<f64>>::haskell_size(tag), 25);
+        let encoded = ParamStruct {
+            a: 1,
+            b: (1.0, 2.0, 3.0),
+        }
+        .try_to_vec()?;
+        assert_eq!(encoded.len(), <ParamStruct<f64>>::haskell_size(tag));
+        Ok(())
+    }
+}
diff --git a/haskell-ffi/src/instances.rs b/haskell-ffi/src/instances.rs
new file mode 100644
index 0000000..beed2b8
--- /dev/null
+++ b/haskell-ffi/src/instances.rs
@@ -0,0 +1,262 @@
+//! ToHaskell and FromHaskell instances for the various standard types mandated
+//! by the [Borsh spec](https://borsh.io/), piggy-backing on the implementation
+//! in the `borsh` crate. The only spec-described types _not_ provided are
+//! user-defined structs and enums.
+
+use borsh::{BorshDeserialize, BorshSerialize};
+use std::{
+    collections::{HashMap, HashSet},
+    hash::Hash,
+    io::{Error, ErrorKind, Write},
+    marker::PhantomData,
+};
+
+use crate::{
+    derive_array_instances, derive_simple_instances, derive_tuple_instances,
+    deriving_via::{tag_ref, untag_val, Haskell},
+    from_haskell::FromHaskell,
+    map_tuple, map_tuple_ref,
+    to_haskell::ToHaskell,
+    HaskellSize,
+};
+
+/*******************************************************************************
+  Simple (non-composite) instances
+*******************************************************************************/
+
+derive_simple_instances!(u8);
+derive_simple_instances!(u16);
+derive_simple_instances!(u32);
+derive_simple_instances!(u64);
+derive_simple_instances!(u128);
+derive_simple_instances!(i8);
+derive_simple_instances!(i16);
+derive_simple_instances!(i32);
+derive_simple_instances!(i64);
+derive_simple_instances!(i128);
+derive_simple_instances!(f32);
+derive_simple_instances!(f64);
+derive_simple_instances!(());
+derive_simple_instances!(String);
+
+/*******************************************************************************
+  Array instances
+
+  This is the same set of sizes as supported by borsh.
+*******************************************************************************/
+
+derive_array_instances!(0);
+derive_array_instances!(1);
+derive_array_instances!(2);
+derive_array_instances!(3);
+derive_array_instances!(4);
+derive_array_instances!(5);
+derive_array_instances!(6);
+derive_array_instances!(7);
+derive_array_instances!(8);
+derive_array_instances!(9);
+derive_array_instances!(10);
+derive_array_instances!(11);
+derive_array_instances!(12);
+derive_array_instances!(13);
+derive_array_instances!(14);
+derive_array_instances!(15);
+derive_array_instances!(16);
+derive_array_instances!(17);
+derive_array_instances!(18);
+derive_array_instances!(19);
+derive_array_instances!(20);
+derive_array_instances!(21);
+derive_array_instances!(22);
+derive_array_instances!(23);
+derive_array_instances!(24);
+derive_array_instances!(25);
+derive_array_instances!(26);
+derive_array_instances!(27);
+derive_array_instances!(28);
+derive_array_instances!(29);
+derive_array_instances!(30);
+derive_array_instances!(31);
+derive_array_instances!(32);
+
+derive_array_instances!(64);
+derive_array_instances!(65);
+
+derive_array_instances!(128);
+derive_array_instances!(256);
+derive_array_instances!(512);
+derive_array_instances!(1024);
+derive_array_instances!(2048);
+
+/*******************************************************************************
+  Composite instances
+
+  This is the same set of tuple sizes as supported by `borsh.`
+*******************************************************************************/
+
+derive_tuple_instances!(T0, T1);
+derive_tuple_instances!(T0, T1, T2);
+derive_tuple_instances!(T0, T1, T2, T3);
+derive_tuple_instances!(T0, T1, T2, T3, T4);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15);
+derive_tuple_instances!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16);
+derive_tuple_instances!(
+    T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16, T17
+);
+derive_tuple_instances!(
+    T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16, T17, T18
+);
+derive_tuple_instances!(
+    T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15, T16, T17, T18, T19
+);
+
+/*******************************************************************************
+  Vec
+*******************************************************************************/
+
+impl<Tag, T: ToHaskell<Tag>> ToHaskell<Tag> for Vec<T> {
+    fn to_haskell<W: Write>(&self, writer: &mut W, _: PhantomData<Tag>) -> Result<(), Error> {
+        let tagged: Vec<&Haskell<Tag, T>> = self.iter().map(tag_ref).collect();
+        tagged.serialize(writer)
+    }
+}
+
+impl<Tag, T: FromHaskell<Tag>> FromHaskell<Tag> for Vec<T> {
+    fn from_haskell(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<Self, Error> {
+        let tagged: Vec<Haskell<Tag, T>> = BorshDeserialize::deserialize(buf)?;
+        Ok(tagged.into_iter().map(untag_val).collect())
+    }
+}
+
+/*******************************************************************************
+  HashMap
+*******************************************************************************/
+
+impl<Tag, K, V> ToHaskell<Tag> for HashMap<K, V>
+where
+    K: Eq + PartialOrd + Hash + ToHaskell<Tag>,
+    V: ToHaskell<Tag>,
+{
+    fn to_haskell<W: Write>(&self, writer: &mut W, _: PhantomData<Tag>) -> Result<(), Error> {
+        let tagged: HashMap<&Haskell<Tag, K>, &Haskell<Tag, V>> =
+            self.iter().map(|(k, v)| (tag_ref(k), tag_ref(v))).collect();
+        tagged.serialize(writer)
+    }
+}
+
+impl<Tag, K, V> FromHaskell<Tag> for HashMap<K, V>
+where
+    K: Eq + Hash + FromHaskell<Tag>,
+    V: FromHaskell<Tag>,
+{
+    fn from_haskell(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<Self, Error> {
+        let tagged: HashMap<Haskell<Tag, K>, Haskell<Tag, V>> = BorshDeserialize::deserialize(buf)?;
+        Ok(tagged
+            .into_iter()
+            .map(|(k, v)| (untag_val(k), untag_val(v)))
+            .collect())
+    }
+}
+
+/*******************************************************************************
+  HashSet
+*******************************************************************************/
+
+impl<Tag, T> ToHaskell<Tag> for HashSet<T>
+where
+    T: Eq + PartialOrd + Hash + ToHaskell<Tag>,
+{
+    fn to_haskell<W: Write>(&self, writer: &mut W, _: PhantomData<Tag>) -> Result<(), Error> {
+        let tagged: HashSet<&Haskell<Tag, T>> = self.iter().map(tag_ref).collect();
+        tagged.serialize(writer)
+    }
+}
+
+impl<Tag, T> FromHaskell<Tag> for HashSet<T>
+where
+    T: Eq + Hash + FromHaskell<Tag>,
+{
+    fn from_haskell(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<Self, Error> {
+        let tagged: HashSet<Haskell<Tag, T>> = BorshDeserialize::deserialize(buf)?;
+        Ok(tagged.into_iter().map(untag_val).collect())
+    }
+}
+
+/*******************************************************************************
+  Option
+*******************************************************************************/
+
+impl<Tag, T: ToHaskell<Tag>> ToHaskell<Tag> for Option<T> {
+    fn to_haskell<W: Write>(&self, writer: &mut W, _: PhantomData<Tag>) -> Result<(), Error> {
+        let tagged: Option<&Haskell<Tag, T>> = self.as_ref().map(tag_ref);
+        tagged.serialize(writer)
+    }
+}
+
+impl<Tag, T: FromHaskell<Tag>> FromHaskell<Tag> for Option<T> {
+    fn from_haskell(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<Self, Error> {
+        let tagged: Option<Haskell<Tag, T>> = BorshDeserialize::deserialize(buf)?;
+        Ok(tagged.map(untag_val))
+    }
+}
+
+/*******************************************************************************
+  Result
+
+  `Result` is not explicitly mentioned by the Borsh spec, but it's ubiquitous
+  and so we provide an instance for it, following the standard rule for enum.
+
+  There is no need for an instance of `FromHaskell`, since this is indicating
+  the result of some Rust-side operation.
+*******************************************************************************/
+
+impl<Tag, T: ToHaskell<Tag>, E: ToHaskell<Tag>> ToHaskell<Tag> for Result<T, E> {
+    fn to_haskell<W: Write>(&self, writer: &mut W, _: PhantomData<Tag>) -> Result<(), Error> {
+        let tagged: Result<&Haskell<Tag, T>, &Haskell<Tag, E>> = match self {
+            Ok(t) => Ok(tag_ref(t)),
+            Err(e) => Err(tag_ref(e)),
+        };
+        tagged.serialize(writer)
+    }
+}
+
+/*******************************************************************************
+  Bool
+
+  The Borsh spec does not mention Bool; we encode `true` as 1 and `false` as 0;
+  this matches what the Haskell `borsh` library does.
+*******************************************************************************/
+
+impl<Tag> HaskellSize<Tag> for bool {
+    fn haskell_size(tag: PhantomData<Tag>) -> usize {
+        u8::haskell_size(tag)
+    }
+}
+
+impl<Tag> ToHaskell<Tag> for bool {
+    fn to_haskell<W: Write>(&self, writer: &mut W, tag: PhantomData<Tag>) -> Result<(), Error> {
+        let as_u8: u8 = if *self { 1 } else { 0 };
+        as_u8.to_haskell(writer, tag)
+    }
+}
+
+impl<Tag> FromHaskell<Tag> for bool {
+    fn from_haskell(buf: &mut &[u8], tag: PhantomData<Tag>) -> Result<Self, Error> {
+        let as_u8 = u8::from_haskell(buf, tag)?;
+        match as_u8 {
+            0 => Ok(false),
+            1 => Ok(true),
+            _ => Err(Error::new(ErrorKind::InvalidData, "Invalid bool")),
+        }
+    }
+}
diff --git a/haskell-ffi/src/lib.rs b/haskell-ffi/src/lib.rs
new file mode 100644
index 0000000..be325a3
--- /dev/null
+++ b/haskell-ffi/src/lib.rs
@@ -0,0 +1,16 @@
+#![feature(array_methods)]
+#![feature(trace_macros)]
+
+mod instances;
+mod macros;
+
+pub mod bincode;
+pub mod deriving_via;
+pub mod from_haskell;
+pub mod haskell_size;
+pub mod to_haskell;
+pub mod use_borsh;
+
+pub use from_haskell::FromHaskell;
+pub use haskell_size::HaskellSize;
+pub use to_haskell::ToHaskell;
diff --git a/haskell-ffi/src/macros.rs b/haskell-ffi/src/macros.rs
new file mode 100644
index 0000000..e69d867
--- /dev/null
+++ b/haskell-ffi/src/macros.rs
@@ -0,0 +1,194 @@
+/*******************************************************************************
+  Auxiliary general-purpose macros
+
+  The `map_tuple` macro is adapted from
+  https://stackoverflow.com/questions/66396814/generating-tuple-indices-based-on-macro-rules-repetition-expansion .
+*******************************************************************************/
+
+/// Map function across all elements of a tuple
+///
+/// ```ignore
+/// map_tuple( [T0, T1], tuple, f )
+/// ```
+///
+/// will become
+///
+/// ```ignore
+/// ( f(tuple.0) , f(tuple.1) )
+/// ```
+///
+/// See also `map_tuple_ref`.
+#[macro_export]
+macro_rules! map_tuple {
+    // Base-case: we are done. Return the accumulator
+    //
+    // We explicitly allow the list of indices to be non-empty (not all indices might be used)
+    ( @, $tuple:ident, $fn:ident, [], [ $($ixs:tt)* ], [ $($acc:tt)* ] ) => {
+        ( $($acc),* )
+    };
+
+    // Recursive-case: add entry to accumulator
+    ( @, $tuple:ident, $fn:ident, [ $t:ident $(,$ts:ident)* ], [ ($ix:tt) $($ixs:tt)* ], [ $($acc:tt)* ] ) => {
+        map_tuple!(@, $tuple, $fn, [ $($ts),* ], [ $($ixs)* ], [ $($acc)* ($fn($tuple . $ix)) ])
+    };
+
+    // Entry-point into the macro
+    ( [ $($ts:ident),* ], $tuple:ident, $fn:ident ) => {
+      map_tuple!(@, $tuple, $fn,
+          // Pass original list of identifiers (only used to determine tuple length)
+          [ $($ts),* ]
+
+          // Pre-defined list of tuple indices
+        , [(0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19)]
+
+          // Empty accumulator
+        , []
+        )
+    }
+}
+
+/// Variation on `map_tuple` that uses a _reference_ to a tuple
+///
+/// TODO: It seems I cannot unify these two macros, because `&self.0` and `(&self).0` are not
+/// equivalent expressions. Is that true..?
+#[macro_export]
+macro_rules! map_tuple_ref {
+    // Base-case: we are done. Return the accumulator
+    //
+    // We explicitly allow the list of indices to be non-empty (not all indices might be used)
+    ( @, $tuple:ident, $fn:ident, [], [ $($ixs:tt)* ], [ $($acc:tt)* ] ) => {
+        ( $($acc),* )
+    };
+
+    // Recursive-case: add entry to accumulator
+    ( @, $tuple:ident, $fn:ident, [ $t:ident $(,$ts:ident)* ], [ ($ix:tt) $($ixs:tt)* ], [ $($acc:tt)* ] ) => {
+        map_tuple_ref!(@, $tuple, $fn, [ $($ts),* ], [ $($ixs)* ], [ $($acc)* ($fn(&$tuple . $ix)) ])
+    };
+
+    // Entry-point into the macro
+    ( [ $($ts:ident),* ], $tuple:ident, $fn:ident ) => {
+      map_tuple_ref!(@, $tuple, $fn,
+          // Pass original list of identifiers (only used to determine tuple length)
+          [ $($ts),* ]
+
+          // Pre-defined list of tuple indices
+        , [(0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19)]
+
+          // Empty accumulator
+        , []
+        )
+    }
+}
+
+/// Fold a list of types
+///
+/// ```ignore
+/// fold_types!( [T0, T1], haskell_size, tag, +, 0);
+/// ```
+///
+/// expands to
+///
+/// ```ignore
+/// 0 + <T0>::haskell_size(tag) + <T1>::haskell_size(tag)
+/// ```
+#[macro_export]
+macro_rules! fold_types {
+    // Base-case: we are done. Return the accumulator
+    ( @, $f:ident, $arg:ident, $op:tt, [], $acc:tt ) => {
+        $acc
+    };
+
+    // Recursive-case: add entry to accumulator
+    ( @, $f:ident, $arg:ident, $op:tt, [ $t:ty $(,$ts:ty)* ], $acc:tt ) => {
+        fold_types!(@, $f, $arg, $op, [ $($ts),* ], ( $acc $op (<$t> :: $f($arg)) ))
+    };
+
+    // Entry-point into the macro
+    ( [ $($ts:ty),* ], $f:ident, $arg:ident, $op:tt, $e:tt ) => {
+        fold_types!(@, $f, $arg, $op, [ $($ts),* ], $e)
+    };
+}
+
+/*******************************************************************************
+  Macros for deriving specific kinds of instances
+*******************************************************************************/
+
+/// Derive `ToHaskell` and `FromHaskell` instances for simple types: types with
+/// no type arguments.
+#[macro_export]
+macro_rules! derive_simple_instances {
+    ($t:ty) => {
+        impl<Tag> ToHaskell<Tag> for $t {
+            fn to_haskell<W: Write>(
+                &self,
+                writer: &mut W,
+                _: PhantomData<Tag>,
+            ) -> Result<(), Error> {
+                self.serialize(writer)
+            }
+        }
+
+        impl<Tag> FromHaskell<Tag> for $t {
+            fn from_haskell(buf: &mut &[u8], _tag: PhantomData<Tag>) -> Result<Self, Error> {
+                <$t>::deserialize(buf)
+            }
+        }
+    };
+}
+
+/// Derive `ToHaskell` and `FromHaskell` instances for arrays of the specified size.
+#[macro_export]
+macro_rules! derive_array_instances {
+    ($sz : literal) => {
+        impl<Tag, T: ToHaskell<Tag>> ToHaskell<Tag> for [T; $sz] {
+            fn to_haskell<W: Write>(
+                &self,
+                writer: &mut W,
+                _: PhantomData<Tag>,
+            ) -> Result<(), Error> {
+                let tagged: [&Haskell<Tag, T>; $sz] = self.each_ref().map(tag_ref);
+                tagged.serialize(writer)
+            }
+        }
+
+        impl<Tag, T: FromHaskell<Tag> + Default + Copy> FromHaskell<Tag> for [T; $sz] {
+            fn from_haskell(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<Self, Error> {
+                let tagged: [Haskell<Tag, T>; $sz] = BorshDeserialize::deserialize(buf)?;
+                Ok(tagged.map(untag_val))
+            }
+        }
+    };
+}
+
+/// Derive `ToHaskell` and `FromHaskell` for tuples with the specified number of type arguments
+/// (i.e., for tuples of the specified size).
+#[macro_export]
+macro_rules! derive_tuple_instances {
+    ($($ts:ident),*) => {
+        impl<Tag, $($ts: ToHaskell<Tag> ),* > ToHaskell<Tag> for ( $($ts ),* ) {
+            fn to_haskell<W: Write>(&self, writer: &mut W,_: PhantomData<Tag>) -> Result<(), Error> {
+                let tagged: ( $(&Haskell<Tag, $ts> ),* ) = map_tuple_ref!( [ $($ts),* ], self, tag_ref );
+                tagged.serialize(writer)
+            }
+        }
+
+        impl<Tag, $($ts: FromHaskell<Tag> ),* > FromHaskell<Tag> for ( $($ts ),* ) {
+            fn from_haskell(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<Self, Error> {
+                let tagged: ( $(Haskell<Tag, $ts> ),* ) = BorshDeserialize::deserialize(buf)?;
+                Ok( map_tuple!( [ $($ts),* ], tagged, untag_val ) )
+            }
+        }
+    };
+}
+
+/// Derive `HaskellSize` instance for tuple with the specified type arguments.
+#[macro_export]
+macro_rules! derive_size_tuple_instance {
+    ($($ts:ident),*) => {
+        impl<Tag, $($ts: HaskellSize<Tag> ),* > HaskellSize<Tag> for ( $($ts),* ) {
+            fn haskell_size(tag: PhantomData<Tag>) -> usize {
+                fold_types!( [ $($ts),* ], haskell_size, tag, +, 0)
+            }
+        }
+    };
+}
diff --git a/haskell-ffi/src/tagged.rs b/haskell-ffi/src/tagged.rs
new file mode 100644
index 0000000..11cf416
--- /dev/null
+++ b/haskell-ffi/src/tagged.rs
@@ -0,0 +1,16 @@
+pub mod borsh_instances;
+mod macros;
+
+use std::marker::PhantomData;
+
+pub struct Tagged<Tag, T> {
+    pub value: T,
+    pub tag: PhantomData<Tag>,
+}
+
+pub fn tag<Tag, T>(t: T) -> Tagged<Tag, T> {
+    Tagged {
+        value: t,
+        tag: PhantomData,
+    }
+}
diff --git a/haskell-ffi/src/to_haskell.rs b/haskell-ffi/src/to_haskell.rs
new file mode 100644
index 0000000..da91378
--- /dev/null
+++ b/haskell-ffi/src/to_haskell.rs
@@ -0,0 +1,93 @@
+use std::{
+    io::{Error, Write},
+    marker::PhantomData,
+};
+
+use crate::HaskellSize;
+
+/*******************************************************************************
+  Main class definition
+*******************************************************************************/
+
+// Copied from `borsh`
+const DEFAULT_SERIALIZER_CAPACITY: usize = 1024;
+
+pub trait ToHaskell<Tag> {
+    /// Serialize data to be sent to Haskell
+    ///
+    /// This is the analogue of `BorshSerialize::serialize`.
+    ///
+    /// The `tag` argument allows client libraries to define additional
+    /// instances of `ToHaskell` for foreign (non-local) types. For example, the
+    /// `solana-sdk-haskell` library can define a `ToHaskell` instance for
+    /// `Keypair`, defined in `solana-sdk`, as long as it uses a tag `Solana`
+    /// defined locally in the `solana-haskell-sdk` package.
+    fn to_haskell<W: Write>(&self, writer: &mut W, tag: PhantomData<Tag>) -> Result<(), Error>;
+
+    fn to_haskell_vec(&self, tag: PhantomData<Tag>) -> Result<Vec<u8>, Error> {
+        let mut result = Vec::with_capacity(DEFAULT_SERIALIZER_CAPACITY);
+        self.to_haskell(&mut result, tag)?;
+        Ok(result)
+    }
+}
+
+impl<Tag, T: ToHaskell<Tag>> ToHaskell<Tag> for &T {
+    fn to_haskell<W: Write>(&self, writer: &mut W, tag: PhantomData<Tag>) -> Result<(), Error> {
+        (*self).to_haskell(writer, tag)
+    }
+}
+
+/*******************************************************************************
+  Derived functionality
+
+  These functions are not defined in the trait itself, to make it clear that
+  they only exist at top-level calls, and will not be recursively called
+  in various `ToHaskell` instances. This is important, because the `len`
+  parameter that gives the length of the buffer only applies to the _overall_
+  buffer.
+*******************************************************************************/
+
+/// Marshall value with fixed-sized encoding
+///
+/// The `out_len` parameter is only used to verify that the Haskell-side and
+/// the Rust side agree on the length of the encoding.
+pub fn marshall_to_haskell_fixed<Tag, T>(t: &T, out: *mut u8, out_len: usize, tag: PhantomData<Tag>)
+where
+    T: HaskellSize<Tag> + ToHaskell<Tag>,
+{
+    let expected_len: usize = T::haskell_size(tag);
+    if out_len != expected_len {
+        panic!(
+            "marshall_to_haskell_fixed: expected buffer of size {}, but got {}",
+            expected_len, out_len
+        )
+    } else {
+        let mut out_len_copy = out_len;
+        marshall_to_haskell_var(t, out, &mut out_len_copy, tag);
+    }
+}
+
+/// Marshall value with variable-sized encoding
+pub fn marshall_to_haskell_var<Tag, T>(
+    t: &T,
+    out: *mut u8,
+    out_len: &mut usize,
+    tag: PhantomData<Tag>,
+) where
+    T: ToHaskell<Tag>,
+{
+    match t.to_haskell_vec(tag) {
+        Ok(vec) => {
+            let slice: &[u8] = vec.as_ref();
+
+            if slice.len() <= *out_len {
+                unsafe {
+                    std::ptr::copy(slice.as_ptr(), out, slice.len());
+                }
+            }
+
+            *out_len = slice.len();
+        }
+        Err(e) => panic!("{}", e),
+    }
+}
diff --git a/haskell-ffi/src/use_borsh.rs b/haskell-ffi/src/use_borsh.rs
new file mode 100644
index 0000000..04047f1
--- /dev/null
+++ b/haskell-ffi/src/use_borsh.rs
@@ -0,0 +1,54 @@
+use borsh::{BorshDeserialize, BorshSerialize};
+use std::{
+    io::{Error, Write},
+    marker::PhantomData,
+};
+
+use crate::{FromHaskell, ToHaskell};
+
+/// Newtype wrapper for defaulting to `borsh` for `ToHaskell`/`FromHaskell`
+///
+/// `ToHaskell`/`FromHaskell` have instances for types such as `Vec<T>`, but
+/// those instances depend on `ToHaskell`/`FromHaskell` for `T`. This
+/// indirection is not always necessary, and may be expensive. The `UseBorsh`
+/// newtype wrapper can be used to mark values where `ToHaskell`/`FromHaskell`
+/// should just piggy-back on Borsh.
+pub struct UseBorsh<T>(pub T);
+
+pub fn unwrap_use_borsh<T>(use_borsh: UseBorsh<T>) -> T {
+    let UseBorsh(t) = use_borsh;
+    t
+}
+
+pub fn unwrap_use_borsh_ref<T>(use_borsh: &UseBorsh<T>) -> &T {
+    let UseBorsh(t) = use_borsh;
+    t
+}
+
+/*******************************************************************************
+  Forwarding instances
+
+  These instances _define_ the `UseBorsh` type
+*******************************************************************************/
+
+impl<Tag, T: BorshSerialize> ToHaskell<Tag> for UseBorsh<T> {
+    fn to_haskell<W: Write>(&self, writer: &mut W, _: PhantomData<Tag>) -> Result<(), Error> {
+        unwrap_use_borsh_ref(self).serialize(writer)
+    }
+}
+
+impl<Tag, T: BorshDeserialize> FromHaskell<Tag> for UseBorsh<T> {
+    fn from_haskell(buf: &mut &[u8], _: PhantomData<Tag>) -> Result<Self, Error> {
+        T::deserialize(buf).map(UseBorsh)
+    }
+}
+
+/*******************************************************************************
+  Additional standard instances
+*******************************************************************************/
+
+impl<T: AsRef<T>> AsRef<T> for UseBorsh<T> {
+    fn as_ref(&self) -> &T {
+        unwrap_use_borsh_ref(self).as_ref()
+    }
+}
diff --git a/rust-toolchain.toml b/rust-toolchain.toml
new file mode 100644
index 0000000..f116d8a
--- /dev/null
+++ b/rust-toolchain.toml
@@ -0,0 +1,2 @@
+[toolchain]
+channel ="nightly"