diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7e22f44..b9f8fb8 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,18 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.9.0] - 2026-02-10
+
+### Added
+
+- **HalProvider interface** for direct HAL device/queue access ([gg#95](https://github.com/gogpu/gg/issues/95))
+  - `HalDevice() any` — returns underlying HAL device for direct GPU access
+  - `HalQueue() any` — returns underlying HAL queue for direct GPU access
+  - Optional interface — use type assertion on DeviceProvider:
+    `if hp, ok := provider.(gpucontext.HalProvider); ok { ... }`
+  - Enables GPU accelerators (e.g., gg SDF pipeline) to share devices with host applications
+    without creating their own wgpu instance
+
 ## [0.8.0] - 2026-02-06
 
 ### Added
@@ -93,6 +105,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **TouchCancelled → TouchCanceled** — US English spelling (misspell linter)
 - Removed unused `DeviceHandle` alias
 
+[0.9.0]: https://github.com/gogpu/gpucontext/releases/tag/v0.9.0
 [0.8.0]: https://github.com/gogpu/gpucontext/releases/tag/v0.8.0
 [0.7.0]: https://github.com/gogpu/gpucontext/releases/tag/v0.7.0
 [0.6.0]: https://github.com/gogpu/gpucontext/releases/tag/v0.6.0
diff --git a/README.md b/README.md
index 5ba0594..96a842d 100644
--- a/README.md
+++ b/README.md
@@ -26,6 +26,7 @@ go get github.com/gogpu/gpucontext
 ## Features
 
 - **DeviceProvider** — Interface for injecting GPU device and queue
+- **HalProvider** — Optional direct access to HAL device/queue for GPU accelerators
 - **WindowProvider** — Window geometry, DPI scale factor, and redraw requests
 - **PlatformProvider** — Clipboard, cursor, dark mode, and accessibility preferences
 - **CursorShape** — 12 standard cursor shapes (arrow, pointer, text, resize, etc.)
@@ -65,6 +66,26 @@ func NewGPUCanvas(provider gpucontext.DeviceProvider) *Canvas {
 }
 ```
 
+### HalProvider (for GPU accelerators)
+
+`HalProvider` is an optional interface on `DeviceProvider` that exposes low-level HAL types.
+This enables GPU accelerators (like gg's SDF pipeline) to share the host device without creating their own:
+
+```go
+// In gogpu/gg - GPU accelerator checks for HAL access
+func (a *SDFAccelerator) SetDeviceProvider(dp gpucontext.DeviceProvider) {
+    if hp, ok := dp.(gpucontext.HalProvider); ok {
+        device := hp.HalDevice().(hal.Device)
+        queue := hp.HalQueue().(hal.Queue)
+        a.initWithSharedDevice(device, queue)
+    }
+}
+
+// In gogpu/gogpu - implements HalProvider
+func (app *App) HalDevice() any { return app.halDevice }
+func (app *App) HalQueue() any  { return app.halQueue }
+```
+
 ### WindowProvider (for UI frameworks)
 
 The `WindowProvider` interface enables UI frameworks to query window dimensions and DPI:
@@ -261,9 +282,9 @@ names := backends.Available() // ["vulkan", "software"]
                           ▼
                    gpucontext
                   (imports gputypes)
-          DeviceProvider, WindowProvider,
-          PlatformProvider, EventSource,
-          Texture, PointerEventSource, Registry
+          DeviceProvider, HalProvider,
+          WindowProvider, PlatformProvider,
+          EventSource, Texture, Registry
                           │
           ┌───────────────┼───────────────┐
           │               │               │
diff --git a/hal_provider.go b/hal_provider.go
new file mode 100644
index 0000000..9650925
--- /dev/null
+++ b/hal_provider.go
@@ -0,0 +1,27 @@
+// Copyright 2026 The gogpu Authors
+// SPDX-License-Identifier: MIT
+
+package gpucontext
+
+// HalProvider is an optional interface for DeviceProviders that can expose
+// low-level HAL types directly. This enables GPU accelerators to share
+// devices without creating their own.
+//
+// The returned any values are hal.Device and hal.Queue from wgpu/hal.
+// Consumers type-assert to the concrete hal types they need.
+//
+// Example usage:
+//
+//	if hp, ok := provider.(gpucontext.HalProvider); ok {
+//	    device := hp.HalDevice().(hal.Device)
+//	    queue := hp.HalQueue().(hal.Queue)
+//	}
+type HalProvider interface {
+	// HalDevice returns the underlying HAL device for direct GPU access.
+	// Returns nil if HAL access is not available.
+	HalDevice() any
+
+	// HalQueue returns the underlying HAL queue for direct GPU access.
+	// Returns nil if HAL access is not available.
+	HalQueue() any
+}