diff --git a/src/crystal/system/thread_local.cr b/src/crystal/system/thread_local.cr
new file mode 100644
index 000000000000..5b7242df12ca
--- /dev/null
+++ b/src/crystal/system/thread_local.cr
@@ -0,0 +1,28 @@
+class Thread
+  struct Local(T)
+    # def initialize
+    #   {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+    # end
+
+    # def initialize(&destructor : Proc(T, Nil))
+    #   {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+    # end
+
+    def get(& : -> T) : T
+      get? || set(yield)
+    end
+
+    # def get? : T?
+    # def set(value : T) : T
+  end
+end
+
+{% if flag?(:wasi) %}
+  require "./wasi/thread_local"
+{% elsif flag?(:unix) %}
+  require "./unix/thread_local"
+{% elsif flag?(:win32) %}
+  require "./win32/thread_local"
+{% else %}
+  {% raise "Thread not supported" %}
+{% end %}
diff --git a/src/crystal/system/unix/thread_local.cr b/src/crystal/system/unix/thread_local.cr
new file mode 100644
index 000000000000..f6bedd279fd1
--- /dev/null
+++ b/src/crystal/system/unix/thread_local.cr
@@ -0,0 +1,30 @@
+require "c/pthread"
+
+class Thread
+  struct Local(T)
+    @key : LibC::PthreadKeyT
+
+    def initialize
+      {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+      err = LibC.pthread_key_create(out @key, nil)
+      raise RuntimeError.from_os_error("pthread_key_create", Errno.new(err)) unless err == 0
+    end
+
+    def initialize(&destructor : Proc(T, Nil))
+      {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+      err = LibC.pthread_key_create(out @key, destructor.unsafe_as(Proc(Void*, Nil)))
+      raise RuntimeError.from_os_error("pthread_key_create", Errno.new(err)) unless err == 0
+    end
+
+    def get? : T?
+      pointer = LibC.pthread_getspecific(@key)
+      pointer.as(T) if pointer
+    end
+
+    def set(value : T) : T
+      err = LibC.pthread_setspecific(@key, value.as(Void*))
+      raise RuntimeError.from_os_error("pthread_setspecific", Errno.new(err)) unless err == 0
+      value
+    end
+  end
+end
diff --git a/src/crystal/system/wasi/thread_local.cr b/src/crystal/system/wasi/thread_local.cr
new file mode 100644
index 000000000000..f1e742780370
--- /dev/null
+++ b/src/crystal/system/wasi/thread_local.cr
@@ -0,0 +1,21 @@
+class Thread
+  struct Local(T)
+    @value : T?
+
+    def initialize
+      {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+    end
+
+    def initialize(&destructor : T ->)
+      {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+    end
+
+    def get? : T?
+      @value
+    end
+
+    def set(value : T) : T
+      @value = value
+    end
+  end
+end
diff --git a/src/crystal/system/win32/thread_local.cr b/src/crystal/system/win32/thread_local.cr
new file mode 100644
index 000000000000..ce53d84a2a29
--- /dev/null
+++ b/src/crystal/system/win32/thread_local.cr
@@ -0,0 +1,30 @@
+require "c/fibersapi"
+
+class Thread
+  struct Local(T)
+    @key : LibC::DWORD
+
+    def initialize
+      {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+      @key = LibC.FlsAlloc
+      raise RuntimeError.from_winerror("FlsAlloc: out of indexes") if @key == LibC::FLS_OUT_OF_INDEXES
+    end
+
+    def initialize(&destructor : Proc(T, Nil))
+      {% raise "T must be a Reference or Pointer" unless T < Reference || T < Pointer %}
+      @key = LibC.FlsAlloc(destructor.unsafe_as(LibC::FLS_CALLBACK_FUNCTION))
+      raise RuntimeError.from_winerror("FlsAlloc: out of indexes") if @key == LibC::FLS_OUT_OF_INDEXES
+    end
+
+    def get? : T?
+      pointer = LibC.FlsGetValue(@key)
+      pointer.as(T) if pointer
+    end
+
+    def set(value : T) : T
+      ret = LibC.FlsSetValue(@key, value.as(Void*))
+      raise RuntimeError.from_winerror("FlsSetValue") if ret == 0
+      value
+    end
+  end
+end
diff --git a/src/lib_c/aarch64-android/c/pthread.cr b/src/lib_c/aarch64-android/c/pthread.cr
index fbab59bd07d7..bd4390df5035 100644
--- a/src/lib_c/aarch64-android/c/pthread.cr
+++ b/src/lib_c/aarch64-android/c/pthread.cr
@@ -26,6 +26,7 @@ lib LibC
   fun pthread_getspecific(__key : PthreadKeyT) : Void*
   fun pthread_join(__pthread : PthreadT, __return_value_ptr : Void**) : Int
   fun pthread_key_create(__key_ptr : PthreadKeyT*, __key_destructor : Void* ->) : Int
+  fun pthread_key_delete(__key_ptr : PthreadKeyT) : Int
 
   fun pthread_mutexattr_destroy(__attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(__attr : PthreadMutexattrT*) : Int
diff --git a/src/lib_c/aarch64-darwin/c/pthread.cr b/src/lib_c/aarch64-darwin/c/pthread.cr
index d146e227197f..43d240435702 100644
--- a/src/lib_c/aarch64-darwin/c/pthread.cr
+++ b/src/lib_c/aarch64-darwin/c/pthread.cr
@@ -13,9 +13,13 @@ lib LibC
   fun pthread_cond_wait(x0 : PthreadCondT*, x1 : PthreadMutexT*) : Int
   fun pthread_create(x0 : PthreadT*, x1 : PthreadAttrT*, x2 : Void* -> Void*, x3 : Void*) : Int
   fun pthread_detach(x0 : PthreadT) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_get_stackaddr_np(x0 : PthreadT) : Void*
   fun pthread_get_stacksize_np(x0 : PthreadT) : SizeT
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -26,4 +30,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/aarch64-darwin/c/sys/types.cr b/src/lib_c/aarch64-darwin/c/sys/types.cr
index 836ac0c1c43d..a0dbc91bdfaa 100644
--- a/src/lib_c/aarch64-darwin/c/sys/types.cr
+++ b/src/lib_c/aarch64-darwin/c/sys/types.cr
@@ -40,6 +40,7 @@ lib LibC
   end
 
   type PthreadT = Void*
+  alias PthreadKeyT = UInt
   alias SSizeT = Long
   alias SusecondsT = Int
   alias TimeT = Long
diff --git a/src/lib_c/aarch64-linux-gnu/c/pthread.cr b/src/lib_c/aarch64-linux-gnu/c/pthread.cr
index 5d9f7b94e225..e3fe03d237b5 100644
--- a/src/lib_c/aarch64-linux-gnu/c/pthread.cr
+++ b/src/lib_c/aarch64-linux-gnu/c/pthread.cr
@@ -25,8 +25,12 @@ lib LibC
   fun pthread_create(newthread : PthreadT*, attr : PthreadAttrT*, start_routine : Void* -> Void*, arg : Void*) : Int
   fun pthread_detach(th : PthreadT) : Int
   fun pthread_getattr_np(thread : PthreadT, attr : PthreadAttrT*) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_equal(thread1 : PthreadT, thread2 : PthreadT) : Int
   fun pthread_join(th : PthreadT, thread_return : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(attr : PthreadMutexattrT*, type : Int) : Int
@@ -37,4 +41,5 @@ lib LibC
   fun pthread_mutex_unlock(mutex : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/aarch64-linux-gnu/c/sys/types.cr b/src/lib_c/aarch64-linux-gnu/c/sys/types.cr
index bef2498d1494..49aeefee2e39 100644
--- a/src/lib_c/aarch64-linux-gnu/c/sys/types.cr
+++ b/src/lib_c/aarch64-linux-gnu/c/sys/types.cr
@@ -41,6 +41,7 @@ lib LibC
   end
 
   alias PthreadT = ULong
+  alias PthreadKeyT = UInt
   alias SSizeT = Long
   alias SusecondsT = Long
   alias TimeT = Long
diff --git a/src/lib_c/aarch64-linux-musl/c/pthread.cr b/src/lib_c/aarch64-linux-musl/c/pthread.cr
index 6ee1a20d90a6..6704cd4c9802 100644
--- a/src/lib_c/aarch64-linux-musl/c/pthread.cr
+++ b/src/lib_c/aarch64-linux-musl/c/pthread.cr
@@ -17,7 +17,11 @@ lib LibC
   fun pthread_create(x0 : PthreadT*, x1 : PthreadAttrT*, x2 : Void* -> Void*, x3 : Void*) : Int
   fun pthread_detach(x0 : PthreadT) : Int
   fun pthread_getattr_np(x0 : PthreadT, x1 : PthreadAttrT*) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -28,4 +32,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/aarch64-linux-musl/c/sys/types.cr b/src/lib_c/aarch64-linux-musl/c/sys/types.cr
index f828ffacffd0..0959f27fc863 100644
--- a/src/lib_c/aarch64-linux-musl/c/sys/types.cr
+++ b/src/lib_c/aarch64-linux-musl/c/sys/types.cr
@@ -54,6 +54,7 @@ lib LibC
   end
 
   type PthreadT = Void*
+  alias PthreadKeyT = UInt
   alias SSizeT = Long
   alias SusecondsT = Long
   alias TimeT = Long
diff --git a/src/lib_c/arm-linux-gnueabihf/c/pthread.cr b/src/lib_c/arm-linux-gnueabihf/c/pthread.cr
index 5d9f7b94e225..e3fe03d237b5 100644
--- a/src/lib_c/arm-linux-gnueabihf/c/pthread.cr
+++ b/src/lib_c/arm-linux-gnueabihf/c/pthread.cr
@@ -25,8 +25,12 @@ lib LibC
   fun pthread_create(newthread : PthreadT*, attr : PthreadAttrT*, start_routine : Void* -> Void*, arg : Void*) : Int
   fun pthread_detach(th : PthreadT) : Int
   fun pthread_getattr_np(thread : PthreadT, attr : PthreadAttrT*) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_equal(thread1 : PthreadT, thread2 : PthreadT) : Int
   fun pthread_join(th : PthreadT, thread_return : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(attr : PthreadMutexattrT*, type : Int) : Int
@@ -37,4 +41,5 @@ lib LibC
   fun pthread_mutex_unlock(mutex : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/arm-linux-gnueabihf/c/sys/types.cr b/src/lib_c/arm-linux-gnueabihf/c/sys/types.cr
index 7a1618b91cf2..a4728582034f 100644
--- a/src/lib_c/arm-linux-gnueabihf/c/sys/types.cr
+++ b/src/lib_c/arm-linux-gnueabihf/c/sys/types.cr
@@ -54,6 +54,7 @@ lib LibC
   end
 
   alias PthreadT = ULong
+  alias PthreadKeyT = UInt
   alias SSizeT = Int
   alias SusecondsT = Long
   alias TimeT = Long
diff --git a/src/lib_c/i386-linux-gnu/c/pthread.cr b/src/lib_c/i386-linux-gnu/c/pthread.cr
index 33274a6cf51d..a621e19acc72 100644
--- a/src/lib_c/i386-linux-gnu/c/pthread.cr
+++ b/src/lib_c/i386-linux-gnu/c/pthread.cr
@@ -25,7 +25,11 @@ lib LibC
   fun pthread_create(newthread : PthreadT*, attr : PthreadAttrT*, start_routine : Void* -> Void*, arg : Void*) : Int
   fun pthread_detach(th : PthreadT) : Int
   fun pthread_getattr_np(thread : PthreadT, attr : PthreadAttrT*) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(th : PthreadT, thread_return : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(attr : PthreadMutexattrT*, type : Int) : Int
@@ -36,4 +40,5 @@ lib LibC
   fun pthread_mutex_unlock(mutex : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/i386-linux-gnu/c/sys/types.cr b/src/lib_c/i386-linux-gnu/c/sys/types.cr
index a1d00215d55d..c38fcb92f102 100644
--- a/src/lib_c/i386-linux-gnu/c/sys/types.cr
+++ b/src/lib_c/i386-linux-gnu/c/sys/types.cr
@@ -41,6 +41,7 @@ lib LibC
   end
 
   alias PthreadT = ULong
+  alias PthreadKeyT = UInt
   alias SSizeT = Int
   alias SusecondsT = Long
   alias TimeT = Long
diff --git a/src/lib_c/i386-linux-musl/c/pthread.cr b/src/lib_c/i386-linux-musl/c/pthread.cr
index 6ee1a20d90a6..6704cd4c9802 100644
--- a/src/lib_c/i386-linux-musl/c/pthread.cr
+++ b/src/lib_c/i386-linux-musl/c/pthread.cr
@@ -17,7 +17,11 @@ lib LibC
   fun pthread_create(x0 : PthreadT*, x1 : PthreadAttrT*, x2 : Void* -> Void*, x3 : Void*) : Int
   fun pthread_detach(x0 : PthreadT) : Int
   fun pthread_getattr_np(x0 : PthreadT, x1 : PthreadAttrT*) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -28,4 +32,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/i386-linux-musl/c/sys/types.cr b/src/lib_c/i386-linux-musl/c/sys/types.cr
index e8a6362b52fe..7e4448048fe3 100644
--- a/src/lib_c/i386-linux-musl/c/sys/types.cr
+++ b/src/lib_c/i386-linux-musl/c/sys/types.cr
@@ -54,6 +54,7 @@ lib LibC
   end
 
   type PthreadT = Void*
+  alias PthreadKeyT = UInt
   alias SSizeT = Int
   alias SusecondsT = Long
   alias TimeT = Long
diff --git a/src/lib_c/x86_64-darwin/c/pthread.cr b/src/lib_c/x86_64-darwin/c/pthread.cr
index d146e227197f..07d973e2ec28 100644
--- a/src/lib_c/x86_64-darwin/c/pthread.cr
+++ b/src/lib_c/x86_64-darwin/c/pthread.cr
@@ -15,7 +15,11 @@ lib LibC
   fun pthread_detach(x0 : PthreadT) : Int
   fun pthread_get_stackaddr_np(x0 : PthreadT) : Void*
   fun pthread_get_stacksize_np(x0 : PthreadT) : SizeT
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -26,4 +30,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/x86_64-darwin/c/sys/types.cr b/src/lib_c/x86_64-darwin/c/sys/types.cr
index 836ac0c1c43d..a0dbc91bdfaa 100644
--- a/src/lib_c/x86_64-darwin/c/sys/types.cr
+++ b/src/lib_c/x86_64-darwin/c/sys/types.cr
@@ -40,6 +40,7 @@ lib LibC
   end
 
   type PthreadT = Void*
+  alias PthreadKeyT = UInt
   alias SSizeT = Long
   alias SusecondsT = Int
   alias TimeT = Long
diff --git a/src/lib_c/x86_64-dragonfly/c/pthread.cr b/src/lib_c/x86_64-dragonfly/c/pthread.cr
index 90bf0c6285a0..17e2d7f94e7c 100644
--- a/src/lib_c/x86_64-dragonfly/c/pthread.cr
+++ b/src/lib_c/x86_64-dragonfly/c/pthread.cr
@@ -19,7 +19,11 @@ lib LibC
   fun pthread_cond_wait(x0 : PthreadCondT*, x1 : PthreadMutexT*) : Int
   fun pthread_create(x0 : PthreadT*, x1 : PthreadAttrT*, x2 : Void* -> Void*, x3 : Void*) : Int
   fun pthread_detach(x0 : PthreadT) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -30,4 +34,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/x86_64-dragonfly/c/sys/types.cr b/src/lib_c/x86_64-dragonfly/c/sys/types.cr
index effbe1227c7f..d5c47dd4eae5 100644
--- a/src/lib_c/x86_64-dragonfly/c/sys/types.cr
+++ b/src/lib_c/x86_64-dragonfly/c/sys/types.cr
@@ -17,6 +17,7 @@ lib LibC
   type PthreadAttrT = Void*
   type PthreadCondT = Void*
   type PthreadCondattrT = Void*
+  alias PthreadKeyT = Int
   type PthreadMutexT = Void*
   type PthreadMutexattrT = Void*
   type PthreadT = Void*
diff --git a/src/lib_c/x86_64-freebsd/c/pthread.cr b/src/lib_c/x86_64-freebsd/c/pthread.cr
index 9cc78c2f2850..f1885a2a1400 100644
--- a/src/lib_c/x86_64-freebsd/c/pthread.cr
+++ b/src/lib_c/x86_64-freebsd/c/pthread.cr
@@ -19,7 +19,11 @@ lib LibC
   fun pthread_cond_wait(x0 : PthreadCondT*, x1 : PthreadMutexT*) : Int
   fun pthread_create(x0 : PthreadT*, x1 : PthreadAttrT*, x2 : Void* -> Void*, x3 : Void*) : Int
   fun pthread_detach(x0 : PthreadT) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -30,4 +34,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_set_name_np(PthreadT, Char*)
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/x86_64-freebsd/c/sys/types.cr b/src/lib_c/x86_64-freebsd/c/sys/types.cr
index 7e0ffad4cd28..77de36fbe71b 100644
--- a/src/lib_c/x86_64-freebsd/c/sys/types.cr
+++ b/src/lib_c/x86_64-freebsd/c/sys/types.cr
@@ -17,6 +17,7 @@ lib LibC
   type PthreadAttrT = Void*
   type PthreadCondT = Void*
   type PthreadCondattrT = Void*
+  alias PthreadKeyT = Int
   type PthreadMutexT = Void*
   type PthreadMutexattrT = Void*
   type PthreadT = Void*
diff --git a/src/lib_c/x86_64-linux-gnu/c/pthread.cr b/src/lib_c/x86_64-linux-gnu/c/pthread.cr
index 33274a6cf51d..a621e19acc72 100644
--- a/src/lib_c/x86_64-linux-gnu/c/pthread.cr
+++ b/src/lib_c/x86_64-linux-gnu/c/pthread.cr
@@ -25,7 +25,11 @@ lib LibC
   fun pthread_create(newthread : PthreadT*, attr : PthreadAttrT*, start_routine : Void* -> Void*, arg : Void*) : Int
   fun pthread_detach(th : PthreadT) : Int
   fun pthread_getattr_np(thread : PthreadT, attr : PthreadAttrT*) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(th : PthreadT, thread_return : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(attr : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(attr : PthreadMutexattrT*, type : Int) : Int
@@ -36,4 +40,5 @@ lib LibC
   fun pthread_mutex_unlock(mutex : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/x86_64-linux-gnu/c/sys/types.cr b/src/lib_c/x86_64-linux-gnu/c/sys/types.cr
index 0fa18856ed06..08faf707a99a 100644
--- a/src/lib_c/x86_64-linux-gnu/c/sys/types.cr
+++ b/src/lib_c/x86_64-linux-gnu/c/sys/types.cr
@@ -40,6 +40,7 @@ lib LibC
     __align : Int
   end
 
+  alias PthreadKeyT = UInt
   alias PthreadT = ULong
   alias SSizeT = Long
   alias SusecondsT = Long
diff --git a/src/lib_c/x86_64-linux-musl/c/pthread.cr b/src/lib_c/x86_64-linux-musl/c/pthread.cr
index 6ee1a20d90a6..6704cd4c9802 100644
--- a/src/lib_c/x86_64-linux-musl/c/pthread.cr
+++ b/src/lib_c/x86_64-linux-musl/c/pthread.cr
@@ -17,7 +17,11 @@ lib LibC
   fun pthread_create(x0 : PthreadT*, x1 : PthreadAttrT*, x2 : Void* -> Void*, x3 : Void*) : Int
   fun pthread_detach(x0 : PthreadT) : Int
   fun pthread_getattr_np(x0 : PthreadT, x1 : PthreadAttrT*) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -28,4 +32,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/x86_64-linux-musl/c/sys/types.cr b/src/lib_c/x86_64-linux-musl/c/sys/types.cr
index 7b05d631a549..60697e828056 100644
--- a/src/lib_c/x86_64-linux-musl/c/sys/types.cr
+++ b/src/lib_c/x86_64-linux-musl/c/sys/types.cr
@@ -54,6 +54,7 @@ lib LibC
   end
 
   type PthreadT = Void*
+  alias PthreadKeyT = UInt
   alias SSizeT = Long
   alias SusecondsT = Long
   alias TimeT = Long
diff --git a/src/lib_c/x86_64-netbsd/c/pthread.cr b/src/lib_c/x86_64-netbsd/c/pthread.cr
index e0f477018777..7f5479bcfbeb 100644
--- a/src/lib_c/x86_64-netbsd/c/pthread.cr
+++ b/src/lib_c/x86_64-netbsd/c/pthread.cr
@@ -25,6 +25,7 @@ lib LibC
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
   alias PthreadKeyDestructor = (Void*) ->
   fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_main_np : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
diff --git a/src/lib_c/x86_64-openbsd/c/pthread.cr b/src/lib_c/x86_64-openbsd/c/pthread.cr
index 15773336f239..b1d4ed2eeed6 100644
--- a/src/lib_c/x86_64-openbsd/c/pthread.cr
+++ b/src/lib_c/x86_64-openbsd/c/pthread.cr
@@ -20,6 +20,7 @@ lib LibC
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
   alias PthreadKeyDestructor = (Void*) ->
   fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_main_np : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
diff --git a/src/lib_c/x86_64-solaris/c/pthread.cr b/src/lib_c/x86_64-solaris/c/pthread.cr
index 538d68263739..cbad672ceea0 100644
--- a/src/lib_c/x86_64-solaris/c/pthread.cr
+++ b/src/lib_c/x86_64-solaris/c/pthread.cr
@@ -19,7 +19,11 @@ lib LibC
   fun pthread_cond_wait(x0 : PthreadCondT*, x1 : PthreadMutexT*) : Int
   fun pthread_create(x0 : PthreadT*, x1 : PthreadAttrT*, x2 : Void* -> Void*, x3 : Void*) : Int
   fun pthread_detach(x0 : PthreadT) : Int
+  fun pthread_getspecific(PthreadKeyT) : Void*
   fun pthread_join(x0 : PthreadT, x1 : Void**) : Int
+  alias PthreadKeyDestructor = (Void*) ->
+  fun pthread_key_create(PthreadKeyT*, PthreadKeyDestructor) : Int
+  fun pthread_key_delete(PthreadKeyT) : Int
   fun pthread_mutexattr_destroy(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_init(x0 : PthreadMutexattrT*) : Int
   fun pthread_mutexattr_settype(x0 : PthreadMutexattrT*, x1 : Int) : Int
@@ -30,4 +34,5 @@ lib LibC
   fun pthread_mutex_unlock(x0 : PthreadMutexT*) : Int
   fun pthread_self : PthreadT
   fun pthread_setname_np(PthreadT, Char*) : Int
+  fun pthread_setspecific(PthreadKeyT, Void*) : Int
 end
diff --git a/src/lib_c/x86_64-solaris/c/sys/types.cr b/src/lib_c/x86_64-solaris/c/sys/types.cr
index b8457bffe861..1054d0204c44 100644
--- a/src/lib_c/x86_64-solaris/c/sys/types.cr
+++ b/src/lib_c/x86_64-solaris/c/sys/types.cr
@@ -45,6 +45,7 @@ lib LibC
   end
 
   alias PthreadT = UInt
+  alias PthreadKeyT = Int
 
   alias SSizeT = Long
   alias SusecondsT = Long
diff --git a/src/lib_c/x86_64-windows-msvc/c/fibersapi.cr b/src/lib_c/x86_64-windows-msvc/c/fibersapi.cr
new file mode 100644
index 000000000000..5b2dfc05a1f9
--- /dev/null
+++ b/src/lib_c/x86_64-windows-msvc/c/fibersapi.cr
@@ -0,0 +1,10 @@
+lib LibC
+  FLS_OUT_OF_INDEXES = 0xFFFFFFFF_u32
+
+  alias FLS_CALLBACK_FUNCTION = Proc(Void*, Nil)
+
+  fun FlsAlloc(FLS_CALLBACK_FUNCTION) : DWORD
+  fun FlsFree(DWORD) : BOOL
+  fun FlsGetValue(DWORD) : Void*
+  fun FlsSetValue(DWORD, Void*) : BOOL
+end
diff --git a/src/regex/lib_pcre2.cr b/src/regex/lib_pcre2.cr
index 6f45a4465219..f6234e30b3c1 100644
--- a/src/regex/lib_pcre2.cr
+++ b/src/regex/lib_pcre2.cr
@@ -264,10 +264,12 @@ lib LibPCRE2
 
   fun jit_stack_create = pcre2_jit_stack_create_8(startsize : LibC::SizeT, maxsize : LibC::SizeT, gcontext : GeneralContext*) : JITStack*
   fun jit_stack_assign = pcre2_jit_stack_assign_8(mcontext : MatchContext*, callable_function : Void* -> JITStack*, callable_data : Void*) : Void
+  fun jit_stack_free = pcre2_jit_stack_free_8(jit_stack : JITStack*) : Void
 
   fun pattern_info = pcre2_pattern_info_8(code : Code*, what : UInt32, where : Void*) : Int
 
   fun match = pcre2_match_8(code : Code*, subject : UInt8*, length : LibC::SizeT, startoffset : LibC::SizeT, options : UInt32, match_data : MatchData*, mcontext : MatchContext*) : Int
+  fun match_data_create = pcre2_match_data_create_8(ovecsize : UInt32, gcontext : GeneralContext*) : MatchData*
   fun match_data_create_from_pattern = pcre2_match_data_create_from_pattern_8(code : Code*, gcontext : GeneralContext*) : MatchData*
   fun match_data_free = pcre2_match_data_free_8(match_data : MatchData*) : Void
 
diff --git a/src/regex/pcre2.cr b/src/regex/pcre2.cr
index b56a4ea68839..ae0b1d823a55 100644
--- a/src/regex/pcre2.cr
+++ b/src/regex/pcre2.cr
@@ -1,5 +1,5 @@
 require "./lib_pcre2"
-require "crystal/thread_local_value"
+require "crystal/system/thread_local"
 
 # :nodoc:
 module Regex::PCRE2
@@ -207,12 +207,14 @@ module Regex::PCRE2
   private def match_impl(str, byte_index, options)
     match_data = match_data(str, byte_index, options) || return
 
-    ovector_count = LibPCRE2.get_ovector_count(match_data)
+    # we reuse the same matchdata allocation, so we must "reimplement" the
+    # behavior of pcre2_match_data_create_from_pattern (get_ovector_count always
+    # returns 65535, aka the maximum):
+    ovector_count = capture_count_impl &+ 1
     ovector = Slice.new(LibPCRE2.get_ovector_pointer(match_data), ovector_count &* 2)
 
     # We need to dup the ovector because `match_data` is re-used for subsequent
-    # matches (see `@match_data`).
-    # Dup brings the ovector data into the realm of the GC.
+    # matches. Dup brings the ovector data into the realm of the GC.
     ovector = ovector.dup
 
     ::Regex::MatchData.new(self, @re, str, byte_index, ovector.to_unsafe, ovector_count.to_i32 &- 1)
@@ -228,43 +230,50 @@ module Regex::PCRE2
 
   class_getter match_context : LibPCRE2::MatchContext* do
     match_context = LibPCRE2.match_context_create(nil)
-    LibPCRE2.jit_stack_assign(match_context, ->(_data) { Regex::PCRE2.jit_stack }, nil)
+    LibPCRE2.jit_stack_assign(match_context, ->(_data) {
+      jit_stack
+    }, nil)
     match_context
   end
 
-  # Returns a JIT stack that's shared in the current thread.
+  # JIT stack is unique per thread.
   #
-  # Only a single `match` function can run per thread at any given time, so there
-  # can't be any concurrent access to the JIT stack.
-  @@jit_stack = Crystal::ThreadLocalValue(LibPCRE2::JITStack*).new
+  # Only a single `match` function can run per thread at any given time, so
+  # there can't be any concurrent access to the JIT stack.
+  @@jit_stack = Thread::Local(LibPCRE2::JITStack*).new do |jit_stack|
+    LibPCRE2.jit_stack_free(jit_stack)
+  end
 
-  def self.jit_stack
+  protected def self.jit_stack
     @@jit_stack.get do
       LibPCRE2.jit_stack_create(32_768, 1_048_576, nil) || raise "Error allocating JIT stack"
     end
   end
 
-  # Match data is shared per instance and thread.
+  # Match data is unique per thread.
   #
-  # Match data contains a buffer for backtracking when matching in interpreted mode (non-JIT).
-  # This buffer is heap-allocated and should be re-used for subsequent matches.
-  @match_data = Crystal::ThreadLocalValue(LibPCRE2::MatchData*).new
+  # Match data contains a buffer for backtracking when matching in interpreted
+  # mode (non-JIT). This buffer is heap-allocated and should be re-used for
+  # subsequent matches.
+  @@match_data = Thread::Local(LibPCRE2::MatchData*).new do |match_data|
+    LibPCRE2.match_data_free(match_data)
+  end
 
-  private def match_data
-    @match_data.get do
-      LibPCRE2.match_data_create_from_pattern(@re, nil)
+  protected def self.match_data : LibPCRE2::MatchData*
+    @@match_data.get do
+      # the ovector size is clamped to 65535 pairs; we declare the maximum
+      # because we allocate the match data buffer once for the thread and need
+      # to adapt to any regular expression
+      LibPCRE2.match_data_create(65_535, nil)
     end
   end
 
   def finalize
-    @match_data.consume_each do |match_data|
-      LibPCRE2.match_data_free(match_data)
-    end
     LibPCRE2.code_free @re
   end
 
   private def match_data(str, byte_index, options)
-    match_data = self.match_data
+    match_data = Regex::PCRE2.match_data
     match_count = LibPCRE2.match(@re, str, str.bytesize, byte_index, pcre2_match_options(options), match_data, PCRE2.match_context)
 
     if match_count < 0