diff --git a/examples/remote/main.c b/examples/remote/main.c
index 81229e173..fc0f93ab1 100644
--- a/examples/remote/main.c
+++ b/examples/remote/main.c
@@ -1,17 +1,28 @@
+#define WGPU_INLINE
+#define WGPU_FUNC
+
 #include "./../../ffi/wgpu-remote.h"
 #include <stdio.h>
 
 int main() {
-    WGPUInfrastructure infra = wgpu_initialize();
+    WGPUInfrastructure infra = wgpu_client_new();
 
-    if (!infra.client || !infra.server || infra.error) {
-        printf("Cannot initialize WGPU: %s", infra.error);
+    if (!infra.client || infra.error) {
+        printf("Cannot initialize WGPU client: %s", infra.error);
+        return 1;
+    }
+
+    WGPUGlobal* server = wgpu_server_new();
+
+    if (!server) {
+        printf("Cannot initialize WGPU client: %s", server);
         return 1;
     }
 
     //TODO: do something meaningful
 
-    wgpu_terminate(infra.client);
+    wgpu_server_delete(server);
+    wgpu_client_delete(infra.client);
 
     return 0;
 }
diff --git a/ffi/wgpu-remote.h b/ffi/wgpu-remote.h
index 555c7dff2..d920eef42 100644
--- a/ffi/wgpu-remote.h
+++ b/ffi/wgpu-remote.h
@@ -14,64 +14,69 @@
 typedef void WGPUEmpty;
 
 
-#include <cstdarg>
-#include <cstdint>
-#include <cstdlib>
-#include <new>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
 
-enum class WGPUPowerPreference {
+typedef enum {
   WGPUPowerPreference_Default = 0,
   WGPUPowerPreference_LowPower = 1,
   WGPUPowerPreference_HighPerformance = 2,
-};
+} WGPUPowerPreference;
 
-template<typename B>
-struct WGPUAdapter;
+typedef struct WGPUClient WGPUClient;
 
-struct WGPUClient;
+typedef struct WGPUGlobal WGPUGlobal;
 
-template<typename B>
-struct WGPUDevice;
+typedef uint64_t WGPUId_Adapter_Dummy;
 
-struct WGPUGlobal;
+typedef WGPUId_Adapter_Dummy WGPUAdapterId;
 
-using WGPUDummy = WGPUEmpty;
+typedef uint64_t WGPUId_Device_Dummy;
 
-template<typename T>
-using WGPUId = uint64_t;
+typedef WGPUId_Device_Dummy WGPUDeviceId;
 
-using WGPUAdapterId = WGPUId<WGPUAdapter<WGPUDummy>>;
-
-using WGPUDeviceId = WGPUId<WGPUDevice<WGPUDummy>>;
-
-struct WGPUInfrastructure {
+typedef struct {
   WGPUClient *client;
   const uint8_t *error;
+} WGPUInfrastructure;
 
-  bool operator==(const WGPUInfrastructure& aOther) const {
-    return client == aOther.client &&
-           error == aOther.error;
-  }
-};
+typedef struct {
+  bool anisotropic_filtering;
+} WGPUExtensions;
 
-using WGPUBackendBit = uint32_t;
+typedef struct {
+  uint32_t max_bind_groups;
+} WGPULimits;
 
-struct WGPURequestAdapterOptions {
+typedef struct {
+  WGPUExtensions extensions;
+  WGPULimits limits;
+} WGPUDeviceDescriptor;
+
+typedef uint32_t WGPUBackendBit;
+
+typedef struct {
   WGPUPowerPreference power_preference;
   WGPUBackendBit backends;
-
-  bool operator==(const WGPURequestAdapterOptions& aOther) const {
-    return power_preference == aOther.power_preference &&
-           backends == aOther.backends;
-  }
-};
-
-extern "C" {
+} WGPURequestAdapterOptions;
 
 WGPU_INLINE
 void wgpu_client_delete(WGPUClient *aClient)
 WGPU_FUNC;
 
+WGPU_INLINE
+void wgpu_client_kill_adapter_ids(const WGPUClient *aClient,
+                                  const WGPUAdapterId *aIds,
+                                  uintptr_t aIdLength)
+WGPU_FUNC;
+
+WGPU_INLINE
+void wgpu_client_kill_device_id(const WGPUClient *aClient,
+                                WGPUDeviceId aId)
+WGPU_FUNC;
+
 WGPU_INLINE
 uintptr_t wgpu_client_make_adapter_ids(const WGPUClient *aClient,
                                        WGPUAdapterId *aIds,
@@ -84,7 +89,14 @@ WGPUDeviceId wgpu_client_make_device_id(const WGPUClient *aClient,
 WGPU_FUNC;
 
 WGPU_INLINE
-WGPUInfrastructure wgpu_client_new()
+WGPUInfrastructure wgpu_client_new(void)
+WGPU_FUNC;
+
+WGPU_INLINE
+void wgpu_server_adapter_request_device(const WGPUGlobal *aGlobal,
+                                        WGPUAdapterId aSelfId,
+                                        const WGPUDeviceDescriptor *aDesc,
+                                        WGPUDeviceId aNewId)
 WGPU_FUNC;
 
 WGPU_INLINE
@@ -92,14 +104,17 @@ void wgpu_server_delete(WGPUGlobal *aGlobal)
 WGPU_FUNC;
 
 WGPU_INLINE
-WGPUGlobal *wgpu_server_new()
+void wgpu_server_device_destroy(const WGPUGlobal *aGlobal,
+                                WGPUDeviceId aSelfId)
 WGPU_FUNC;
 
 WGPU_INLINE
-WGPUAdapterId wgpu_server_request_adapter(const WGPUGlobal *aGlobal,
-                                          const WGPURequestAdapterOptions *aDesc,
-                                          const WGPUAdapterId *aIds,
-                                          uintptr_t aIdLength)
+WGPUAdapterId wgpu_server_instance_request_adapter(const WGPUGlobal *aGlobal,
+                                                   const WGPURequestAdapterOptions *aDesc,
+                                                   const WGPUAdapterId *aIds,
+                                                   uintptr_t aIdLength)
 WGPU_FUNC;
 
-} // extern "C"
+WGPU_INLINE
+WGPUGlobal *wgpu_server_new(void)
+WGPU_FUNC;
diff --git a/wgpu-native/cbindgen.toml b/wgpu-native/cbindgen.toml
index 9e296d5de..f111779e1 100644
--- a/wgpu-native/cbindgen.toml
+++ b/wgpu-native/cbindgen.toml
@@ -32,7 +32,6 @@ bitflags = true
 
 [defines]
 "feature = local" = "WGPU_LOCAL"
-"feature = remote" = "WGPU_REMOTE"
 "feature = gfx-backend-gl" = "WGPU_BACKEND_GL"
 "feature = winit" = "WGPU_WINIT"
 "feature = glutin" = "WGPU_GLUTIN"
diff --git a/wgpu-native/src/command/allocator.rs b/wgpu-native/src/command/allocator.rs
index 07ce1bfe9..fd33c367b 100644
--- a/wgpu-native/src/command/allocator.rs
+++ b/wgpu-native/src/command/allocator.rs
@@ -60,16 +60,6 @@ pub struct CommandAllocator<B: hal::Backend> {
 }
 
 impl<B: GfxBackend> CommandAllocator<B> {
-    pub fn new(queue_family: hal::queue::QueueFamilyId) -> Self {
-        CommandAllocator {
-            queue_family,
-            inner: Mutex::new(Inner {
-                pools: HashMap::new(),
-                pending: Vec::new(),
-            }),
-        }
-    }
-
     pub(crate) fn allocate(
         &self,
         device_id: Stored<DeviceId>,
@@ -103,6 +93,18 @@ impl<B: GfxBackend> CommandAllocator<B> {
             features,
         }
     }
+}
+
+impl<B: hal::Backend> CommandAllocator<B> {
+    pub fn new(queue_family: hal::queue::QueueFamilyId) -> Self {
+        CommandAllocator {
+            queue_family,
+            inner: Mutex::new(Inner {
+                pools: HashMap::new(),
+                pending: Vec::new(),
+            }),
+        }
+    }
 
     pub fn extend(&self, cmd_buf: &CommandBuffer<B>) -> B::CommandBuffer {
         let mut inner = self.inner.lock();
diff --git a/wgpu-native/src/command/mod.rs b/wgpu-native/src/command/mod.rs
index 4313ffe50..efa2d9956 100644
--- a/wgpu-native/src/command/mod.rs
+++ b/wgpu-native/src/command/mod.rs
@@ -112,7 +112,7 @@ pub struct CommandBuffer<B: hal::Backend> {
     pub(crate) raw: Vec<B::CommandBuffer>,
     is_recording: bool,
     recorded_thread_id: ThreadId,
-    device_id: Stored<DeviceId>,
+    pub(crate) device_id: Stored<DeviceId>,
     pub(crate) life_guard: LifeGuard,
     pub(crate) trackers: TrackerSet,
     pub(crate) used_swap_chain: Option<(Stored<TextureViewId>, B::Framebuffer)>,
diff --git a/wgpu-native/src/device.rs b/wgpu-native/src/device.rs
index ac07e7f8c..0510655a8 100644
--- a/wgpu-native/src/device.rs
+++ b/wgpu-native/src/device.rs
@@ -768,6 +768,21 @@ impl<B: GfxBackend> Device<B> {
     }
 }
 
+impl<B: hal::Backend> Device<B> {
+    pub(crate) fn destroy_bind_group(&self, bind_group: binding_model::BindGroup<B>) {
+        unsafe {
+            self.desc_allocator.lock().free(iter::once(bind_group.raw));
+        }
+    }
+
+    pub(crate) fn destroy_self(self) {
+        self.com_allocator.destroy(&self.raw);
+        unsafe {
+            self.desc_allocator.lock().cleanup(&self.raw);
+        }
+    }
+}
+
 #[cfg(feature = "local")]
 #[no_mangle]
 pub extern "C" fn wgpu_device_get_limits(_device_id: DeviceId, limits: &mut Limits) {
diff --git a/wgpu-native/src/hub.rs b/wgpu-native/src/hub.rs
index a51ac6eae..0e21b512d 100644
--- a/wgpu-native/src/hub.rs
+++ b/wgpu-native/src/hub.rs
@@ -389,6 +389,67 @@ impl<B: GfxBackend> Default for Hub<B> {
     }
 }
 
+impl<B: hal::Backend> Drop for Hub<B> {
+    fn drop(&mut self) {
+        use crate::resource::TextureViewInner;
+        use hal::device::Device as _;
+
+        let mut devices = self.devices.data.write();
+
+        for (_, (sampler, _)) in self.samplers.data.write().map.drain() {
+            unsafe {
+                devices[sampler.device_id.value].raw.destroy_sampler(sampler.raw);
+            }
+        }
+        {
+            let textures = self.textures.data.read();
+            for (_, (texture_view, _)) in self.texture_views.data.write().map.drain() {
+                match texture_view.inner {
+                    TextureViewInner::Native { raw, source_id } => {
+                        let device = &devices[textures[source_id.value].device_id.value];
+                        unsafe {
+                            device.raw.destroy_image_view(raw);
+                        }
+                    }
+                    TextureViewInner::SwapChain { .. } => {} //TODO
+                }
+            }
+        }
+        for (_, (texture, _)) in self.textures.data.write().map.drain() {
+            unsafe {
+                devices[texture.device_id.value].raw.destroy_image(texture.raw);
+            }
+        }
+        for (_, (buffer, _)) in self.buffers.data.write().map.drain() {
+            unsafe {
+                devices[buffer.device_id.value].raw.destroy_buffer(buffer.raw);
+            }
+        }
+        for (_, (command_buffer, _)) in self.command_buffers.data.write().map.drain() {
+            devices[command_buffer.device_id.value].com_allocator.after_submit(command_buffer, 0);
+        }
+        for (_, (bind_group, _)) in self.bind_groups.data.write().map.drain() {
+            let device = &devices[bind_group.device_id.value];
+            device.destroy_bind_group(bind_group);
+        }
+
+        //TODO:
+        // self.compute_pipelines
+        // self.compute_passes
+        // self.render_pipelines
+        // self.render_passes
+        // self.bind_group_layouts
+        // self.pipeline_layouts
+        // self.shader_modules
+        // self.swap_chains
+        // self.adapters
+
+        for (_, (device, _)) in devices.map.drain() {
+            device.destroy_self();
+        }
+    }
+}
+
 #[derive(Debug, Default)]
 pub struct Hubs {
     #[cfg(any(
@@ -424,6 +485,16 @@ impl Global {
     pub fn new(name: &str) -> Self {
         Self::new_impl(name)
     }
+
+    #[cfg(not(feature = "local"))]
+    pub fn delete(self) {
+        let Global { mut instance, surfaces, hubs } = self;
+        drop(hubs);
+        // destroy surfaces
+        for (_, (surface, _)) in surfaces.data.write().map.drain() {
+            instance.destroy_surface(surface);
+        }
+    }
 }
 
 #[cfg(feature = "local")]
diff --git a/wgpu-native/src/instance.rs b/wgpu-native/src/instance.rs
index b5b733a6f..ab9e08231 100644
--- a/wgpu-native/src/instance.rs
+++ b/wgpu-native/src/instance.rs
@@ -58,6 +58,31 @@ impl Instance {
             dx11: gfx_backend_dx11::Instance::create(name, version).unwrap(),
         }
     }
+
+    #[cfg(not(feature = "local"))]
+    pub(crate) fn destroy_surface(&mut self, surface: Surface) {
+        //TODO: fill out the proper destruction once we are on gfx-0.4
+        #[cfg(any(
+            not(any(target_os = "ios", target_os = "macos")),
+            feature = "gfx-backend-vulkan"
+        ))]
+        {
+            if let Some(_suf) = surface.vulkan {
+                //self.vulkan.as_mut().unwrap().destroy_surface(suf);
+            }
+        }
+        #[cfg(any(target_os = "ios", target_os = "macos"))]
+        {
+            //self.metal.destroy_surface(surface.metal);
+        }
+        #[cfg(windows)]
+        {
+            if let Some(_suf) = surface.dx12 {
+                //self.dx12.as_mut().unwrap().destroy_surface(suf);
+            }
+            //self.dx11.destroy_surface(surface.dx11);
+        }
+    }
 }
 
 type GfxSurface<B> = <B as hal::Backend>::Surface;
diff --git a/wgpu-remote/cbindgen.toml b/wgpu-remote/cbindgen.toml
index d139f9f2a..2d552f481 100644
--- a/wgpu-remote/cbindgen.toml
+++ b/wgpu-remote/cbindgen.toml
@@ -14,7 +14,7 @@ include_version = true
 braces = "SameLine"
 line_length = 100
 tab_width = 2
-language = "C++"
+language = "C"
 
 [export]
 prefix = "WGPU"
diff --git a/wgpu-remote/src/server.rs b/wgpu-remote/src/server.rs
index a249df84c..719591ee4 100644
--- a/wgpu-remote/src/server.rs
+++ b/wgpu-remote/src/server.rs
@@ -13,8 +13,8 @@ pub extern "C" fn wgpu_server_new() -> *mut wgn::Global {
 #[no_mangle]
 pub extern "C" fn wgpu_server_delete(global: *mut wgn::Global) {
     log::info!("Terminating WGPU server");
-    //TODO: proper cleanup
-    let _ = unsafe { Box::from_raw(global) };
+    unsafe { Box::from_raw(global) }.delete();
+    log::info!("\t...done");
 }
 
 #[no_mangle]
@@ -38,3 +38,9 @@ pub extern "C" fn wgpu_server_adapter_request_device(
     use wgn::adapter_request_device as func;
     wgn::gfx_select!(self_id => func(global, self_id, desc, new_id));
 }
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_device_destroy(global: &wgn::Global, self_id: wgn::DeviceId) {
+    use wgn::device_destroy as func;
+    wgn::gfx_select!(self_id => func(global, self_id))
+}