diff --git a/docs/software/TODO.md b/docs/software/TODO.md
index 2940a5fb..46f943c0 100644
--- a/docs/software/TODO.md
+++ b/docs/software/TODO.md
@@ -10,7 +10,7 @@ For app cleanup:
 * DONE write devapi user guide
 * DONE update android code: https://developer.android.com/topic/libraries/view-binding/migration
 * only do wantReplies once per packet type, if we change network settings force it again
-* make gpio watch work, use thread and setup
+* test GPIO watch
 * DONE make hello world example service
 * make python ping command
 * DONE have python tool check max packet size before sending to device
diff --git a/docs/software/plugin-api.md b/docs/software/plugin-api.md
index ba544928..e0699a6b 100644
--- a/docs/software/plugin-api.md
+++ b/docs/software/plugin-api.md
@@ -49,6 +49,11 @@ The easiest way to get started is:
 * Rebuild with your new messaging goodness and install on the device
 * Use the [meshtastic commandline tool](https://github.com/meshtastic/Meshtastic-python) to send a packet to your board "meshtastic --dest 1234 --ping"
 
+## Threading
+
+It is very common that you would like your plugin to be invoked periodically.
+We use a crude/basic cooperative threading system to allow this on any of our supported platforms.  Simply inherit from OSThread and implement runOnce().  See the OSThread [documentation](/src/concurrency/OSThread.h) for more details.  For an example consumer of this API see RemoteHardwarePlugin::runOnce.
+
 ## Picking a port number
 
 For any new 'apps' that run on the device or via sister apps on phones/PCs they should pick and use a unique 'portnum' for their application.
diff --git a/src/plugins/RemoteHardwarePlugin.cpp b/src/plugins/RemoteHardwarePlugin.cpp
index 11aaad23..dee93363 100644
--- a/src/plugins/RemoteHardwarePlugin.cpp
+++ b/src/plugins/RemoteHardwarePlugin.cpp
@@ -10,6 +10,44 @@ RemoteHardwarePlugin remoteHardwarePlugin;
 
 #define NUM_GPIOS 64
 
+// Because (FIXME) we currently don't tell API clients status on sent messages
+// we need to throttle our sending, so that if a gpio is bouncing up and down we
+// don't generate more messages than the net can send. So we limit watch messages to 
+// a max of one change per 30 seconds
+#define WATCH_INTERVAL_MSEC (30 * 1000)
+
+/// Set pin modes for every set bit in a mask
+static void pinModes(uint64_t mask, uint8_t mode) {
+    for (uint8_t i = 0; i < NUM_GPIOS; i++) {
+        if (mask & (1 << i)) {
+            pinMode(i, mode);
+        }
+    }
+}
+
+/// Read all the pins mentioned in a mask
+static uint64_t digitalReads(uint64_t mask) {
+    uint64_t res = 0;
+
+    pinModes(mask, INPUT_PULLUP);
+
+    for (uint8_t i = 0; i < NUM_GPIOS; i++) {
+        uint64_t m = 1 << i;
+        if (mask & m) {
+            if (digitalRead(i))
+                res |= m;
+        }
+    }
+
+    return res;
+}
+
+
+RemoteHardwarePlugin::RemoteHardwarePlugin()
+    : ProtobufPlugin("remotehardware", PortNum_REMOTE_HARDWARE_APP, HardwareMessage_fields),
+    concurrency::OSThread("remotehardware")
+{
+}
 
 bool RemoteHardwarePlugin::handleReceivedProtobuf(const MeshPacket &req, const HardwareMessage &p)
 {
@@ -22,24 +60,17 @@ bool RemoteHardwarePlugin::handleReceivedProtobuf(const MeshPacket &req, const H
             uint64_t mask = 1 << i;
             if (p.gpio_mask & mask) {
                 digitalWrite(i, (p.gpio_value & mask) ? 1 : 0);
-                pinMode(i, OUTPUT);
             }
         }
+        pinModes(p.gpio_mask, OUTPUT);
+
         break;
-        
+
     case HardwareMessage_Type_READ_GPIOS: {
         // Print notification to LCD screen
         screen->print("Read GPIOs\n");
 
-        uint64_t res = 0;
-        for (uint8_t i = 0; i < NUM_GPIOS; i++) {
-            uint64_t mask = 1 << i;
-            if (p.gpio_mask & mask) {
-                pinMode(i, INPUT_PULLUP);
-                if (digitalRead(i))
-                    res |= (1 << i);
-            }
-        }
+        uint64_t res = digitalReads(p.gpio_mask);
 
         // Send the reply
         HardwareMessage reply = HardwareMessage_init_default;
@@ -51,6 +82,14 @@ bool RemoteHardwarePlugin::handleReceivedProtobuf(const MeshPacket &req, const H
         break;
     }
 
+    case HardwareMessage_Type_WATCH_GPIOS: {
+        watchGpios = p.gpio_mask;
+        lastWatchMsec = 0; // Force a new publish soon
+        previousWatch = ~watchGpios; // generate a 'previous' value which is guaranteed to not match (to force an initial publish)
+        enabled = true; // Let our thread run at least once
+        break;
+    }
+
     case HardwareMessage_Type_READ_GPIOS_REPLY:
     case HardwareMessage_Type_GPIOS_CHANGED:
         break; // Ignore - we might see our own replies
@@ -61,3 +100,30 @@ bool RemoteHardwarePlugin::handleReceivedProtobuf(const MeshPacket &req, const H
     }
     return true; // handled
 }
+
+int32_t RemoteHardwarePlugin::runOnce() {
+    if(watchGpios) {
+        uint32_t now = millis();
+
+        if(now - lastWatchMsec >= WATCH_INTERVAL_MSEC) {
+            uint64_t curVal = digitalReads(watchGpios);
+
+            if(curVal != previousWatch) {
+                previousWatch = curVal;
+
+                // Something changed!  Tell the world with a broadcast message
+                HardwareMessage reply = HardwareMessage_init_default;
+                reply.typ = HardwareMessage_Type_GPIOS_CHANGED;
+                reply.gpio_value = curVal;
+                MeshPacket *p = allocDataProtobuf(reply);
+                service.sendToMesh(p);
+            }
+        }
+    }
+    else {
+        // No longer watching anything - stop using CPU
+        enabled = false;
+    }
+
+    return 200; // Poll our GPIOs every 200ms (FIXME, make adjustable via protobuf arg)
+}
\ No newline at end of file
diff --git a/src/plugins/RemoteHardwarePlugin.h b/src/plugins/RemoteHardwarePlugin.h
index faae4894..fe9bf960 100644
--- a/src/plugins/RemoteHardwarePlugin.h
+++ b/src/plugins/RemoteHardwarePlugin.h
@@ -1,17 +1,26 @@
 #pragma once
 #include "ProtobufPlugin.h"
 #include "remote_hardware.pb.h"
+#include "concurrency/OSThread.h"
 
 /**
  * A plugin that provides easy low-level remote access to device hardware.
  */
-class RemoteHardwarePlugin : public ProtobufPlugin<HardwareMessage>
+class RemoteHardwarePlugin : public ProtobufPlugin<HardwareMessage>, private concurrency::OSThread
 {
+    /// The current set of GPIOs we've been asked to watch for changes
+    uint64_t watchGpios = 0; 
+
+    /// The previously read value of watched pins
+    uint64_t previousWatch = 0;
+
+    /// The timestamp of our last watch event (we throttle watches to 1 change every 30 seconds)
+    uint32_t lastWatchMsec = 0;
   public:
     /** Constructor
      * name is for debugging output
      */
-    RemoteHardwarePlugin() : ProtobufPlugin("remotehardware", PortNum_REMOTE_HARDWARE_APP, HardwareMessage_fields) {}
+    RemoteHardwarePlugin();
 
   protected:
     /** Called to handle a particular incoming message
@@ -19,6 +28,16 @@ class RemoteHardwarePlugin : public ProtobufPlugin<HardwareMessage>
     @return true if you've guaranteed you've handled this message and no other handlers should be considered for it
     */
     virtual bool handleReceivedProtobuf(const MeshPacket &mp, const HardwareMessage &p);
+
+    /**
+     * Periodically read the gpios we have been asked to WATCH, if they have changed,
+     * broadcast a message with the change information.
+     * 
+     * The method that will be called each time our thread gets a chance to run
+     *
+     * Returns desired period for next invocation (or RUN_SAME for no change)
+     */
+    virtual int32_t runOnce();
 };
 
 extern RemoteHardwarePlugin remoteHardwarePlugin;
\ No newline at end of file