diff --git a/guides/sets.v b/guides/sets.v
new file mode 100644
index 0000000000000000000000000000000000000000..33397b405a7e5106b50b8029488484b175167e13
--- /dev/null
+++ b/guides/sets.v
@@ -0,0 +1,99 @@
+From stdpp Require Import gmap sorting.
+
+(* [gset K] is the type of sets you probably want to use. It enjoys good
+properties w.r.t. computation (most operations are logarithmic or linear) and
+equality. It works for any type [K] that is countable (see type class
+[Countable]). There are [Countable] instances for all the usual types, [nat],
+[list nat], [nat * nat], [gset nat], etc. *)
+
+Check gset.
+
+(* All the set operations are overloaded via type classes, but there are
+instances for [gset]. The most important operations are:
+
+- [elem_of], notation [∈]
+- [subseteq], notation [⊆]
+- [subset], notation [⊂]
+- [empty], notation [∅]
+- [singleton], notation [ {[ x ]} ]
+- [union], notation [∪]
+- [intersection], notation [∩]
+- [difference], notation [∖]
+*)
+
+(* Let us try to type check some stuff. *)
+
+Check {[ 10 ]} : gset nat.
+
+Check {[ [10] ]} : gset (list nat).
+
+Check {[ {[ 10 ]} ]} : gset (gset nat).
+
+Check (λ X : gset nat, X ∪ {[ 10 ]}).
+
+(** And let write some lemmas. The most useful tactic is [set_solver]. *)
+
+Lemma some_stuff (X Y Z : gset nat) :
+  X ∪ Y ∩ X ∪ Z ∩ X = (Y ∪ X ∪ Z ∖ ∅ ∪ X) ∩ X.
+Proof. set_solver. Qed.
+
+Lemma some_stuff_poly `{Countable A} (X Y Z : gset nat) :
+  X ∪ Y ∩ X ∪ Z ∩ X = (Y ∪ X ∪ Z ∖ ∅ ∪ X) ∩ X.
+Proof. set_solver. Qed.
+
+(** If you want to search for lemmas, search for the operations, not [gset]
+since all lemmas are overloaded. *)
+
+Search difference intersection.
+
+(** Some important notes:
+
+- The lemmas look a bit dounting because of the additional arguments due to
+  type classes, but these arguments can mostly be ignored
+- There are both lemmas about Leibniz equality [=] and setoid equality [≡].
+  The first ones are suffixed [_L]. For [gset] you always want to use [=] (and
+  thus the [_L] lemmas) because we have [X ≡ Y ↔ X = Y]. This is not the case
+  for other implementations of sets, like [propset A := A → Prop] or
+  [listset A := list A], hence [≡] is useful in the general case. *)
+
+(** Some other examples *)
+
+Definition evens (X : gset nat) : gset nat :=
+  filter (λ x, (2 | x)) X.
+Definition intersection_alt (X Y : gset nat) : gset nat :=
+  filter (.∈ Y) X.
+Definition add_one (X : gset nat) : gset nat :=
+  set_map S X.
+Definition until_n (n : nat) : gset nat :=
+  set_seq 0 n.
+
+(** Keep in mind that [filter], [set_map], [set_seq], etc are overloaded via
+type classes. So, you need sufficient type information in your definitions and
+you won't find much about them when searching for [gset]. *)
+
+(** When computing with sets, always make sure to cast the result that you want
+to display is a simple type like [nat], [list nat], [bool], etc. The result
+of a [gset] computation is a big sigma type with lots of noise, so it won't be
+pretty (or useful) to look at. *)
+
+Eval vm_compute in (elements (add_one (evens {[ 10; 11; 14 ]}))).
+Eval vm_compute in (elements (evens (until_n 40))).
+(** [elements] converts a set to a list. They are not sorted, but you can do
+that yourself. *)
+Eval vm_compute in (merge_sort (≤) (elements (evens (until_n 40)))).
+Eval vm_compute in (fresh ({[ 10; 12 ]} : gset nat)).
+Eval vm_compute in (size ({[ 10; 12 ]} : gset nat)).
+(** You can use [bool_decide] to turn decidable [Prop]s into [bool]s. *)
+Eval vm_compute in bool_decide (10 ∈ evens {[ 10; 11 ]}).
+Eval vm_compute in (bool_decide ({[ 10; 14; 17 ]} ⊂ until_n 40)).
+Eval vm_compute in (bool_decide (set_Forall (λ x, (2 | x)) (evens (until_n 40)))).
+
+(** Want to know more:
+
+- Look up the definitions of the type classes for [union], [intersection], etc.,
+  interfaces [SimpleSet], [Set_], etc. in [theories/base.v].
+- Look up the generic theory of sets in [theories/sets.v].
+- Look up the generic theory of finite sets in [theories/fin_sets.v].
+- Probably don't look into the implementation of [gset] in [theories/gmap.v],
+  unless you are very interested in encodings as bit strings and radix-2 search
+  trees. You should treat [gset] as a black box. *)