diff --git a/docs/usage/modules.md b/docs/usage/modules.md
index 897fc1b3ec..076a4ccc76 100644
--- a/docs/usage/modules.md
+++ b/docs/usage/modules.md
@@ -50,9 +50,11 @@ print(CameraModule.io())
 └┬─────────────┘
  ├─ color_image: Image
  ├─ camera_info: CameraInfo
+ ├─ tf: TFMessage
  │
  ├─ RPC build() -> None
  ├─ RPC get_skills() -> list
+ ├─ RPC peek_stream(stream_name: str, timeout: float) -> Any
  ├─ RPC set_module_ref(name: str, module_ref: RPCClient) -> None
  ├─ RPC set_transport(stream_name: str, transport: Transport) -> bool
  ├─ RPC start() -> None
@@ -139,7 +141,7 @@ from dimos.hardware.sensors.camera.module import CameraModule
 camera = CameraModule()
 detector = Detection2DModule()
 
-detector.image.connect(camera.color_image)
+detector.color_image.connect(camera.color_image)
 
 camera.start()
 detector.start()
@@ -472,3 +474,127 @@ to_svg(agentic, "assets/go2_agentic.svg")
 ![output](assets/go2_agentic.svg)
 
 To see more information on how to use Blueprints, see [Blueprints](/docs/usage/blueprints.md).
+
+## Low level manual plumbing
+
+Most of the time this is a bad idea, but you can peak behind blueprints, autoconnect etc, and deal with pubsub directly at lower and lower layers of abstraction
+
+### Connecting via transports
+
+`.connect` wires two modules living in the same process. Alternatively, you can assign a **transport** a typed pub/sub channel such as LCM or Zenoh. The modules never hold a reference to each other, so they can run as separate scripts.
+
+`camera_script.py`:
+
+```python skip
+import time
+from dimos.core.transport import LCMTransport
+from dimos.hardware.sensors.camera.module import CameraModule
+from dimos.msgs.sensor_msgs.Image import Image
+
+camera = CameraModule()
+camera.color_image.transport = LCMTransport("/camera/rgb", Image)
+
+camera.start()
+time.sleep(10)
+camera.stop()
+```
+
+`detector_script.py`:
+
+```python skip
+import time
+from dimos.core.transport import LCMTransport
+from dimos.msgs.sensor_msgs.Image import Image
+from dimos.perception.detection.module2D import Detection2DModule
+
+detector = Detection2DModule()
+detector.color_image.transport = LCMTransport("/camera/rgb", Image)
+
+detector.start()
+detector.detections.subscribe(print)
+time.sleep(10)
+detector.stop()
+```
+
+Run each in its own terminal and detections start printing as soon as both are up. The channel name and message type must match on both sides `LCMTransport("/camera/rgb", Image)` maps to the typed channel `/camera/rgb#sensor_msgs.Image`, which you can watch with `dimos spy` or `dimos topic echo /camera/rgb`.
+
+Available transports live in `dimos.core.transport` (`LCMTransport`, `ZenohTransport`, `SHMTransport`, ...); see [Transports](/docs/usage/transports/index.md) for choosing between them.
+
+### Raw transports (no modules)
+
+A transport works on its own — init one and send/receive from a plain script, no module or stream declarations needed:
+
+```python skip
+from dimos.core.transport import LCMTransport
+from dimos.msgs.std_msgs.String import String
+
+chat = LCMTransport("/chat", String)
+
+unsubscribe = chat.subscribe(print)     # receive
+chat.publish(String(data="hello"))      # send
+
+msg = chat.get_next()                   # or block for the next message
+chat.stop()
+```
+
+Any other script (or module stream) on the same channel sees the traffic. This is handy for quick probes and debug scripts. For a robot system, prefer modules so the streams show up in blueprints and introspection.
+
+### Dynamic streams
+
+Our system is very flexible, but this is dangerous as you lose essentially all support above pubsub. Autoconnect, blueprints, configuration support (dimos --transport=...) etc
+
+Streams don't have to be class annotations, a module can grow inputs and outputs at runtime, and they can even be attached externally.
+Construct the stream and assign it as an attribute.
+
+module has `inputs`/`outputs`/`io()` attributes and is a ble to discover streams attached.
+
+```python ansi=false
+from dimos.core.module import Module
+from dimos.core.stream import In, Out
+from dimos.core.transport import LCMTransport
+from dimos.msgs.std_msgs.String import String
+
+class Dyn(Module):
+    def start(self):
+        # module can add a random input
+        self.echo = In(String, "echo", m)
+        print("Externally attached output:", self.words)
+
+m = Dyn()
+
+m.words = Out(String, "words", m)
+m.start()
+
+
+m.words.transport = LCMTransport("/words", String)
+m.echo.transport = LCMTransport("/words", String)
+
+
+print("\nInputs:")
+print(m.inputs)
+print(m.inputs['echo'])
+print("\nOutputs:")
+print(m.outputs)
+print(m.outputs['words'])
+
+print("\Send/Receive Test:")
+
+# we can subscribe to topics from anywhere also
+m.echo.subscribe(print)
+m.words.publish(String(data="loopback over LCM"))
+
+```
+
+```results
+Externally attached output: Out words[String] @ Dyn
+
+Inputs:
+{'echo': <dimos.core.stream.In object at 0x7f03b4bb32c0>}
+In echo[String] @ Dyn via LCMTransport(/words#std_msgs.String)
+
+Outputs:
+{'words': <dimos.core.stream.Out object at 0x7f03b4b7f860>}
+Out words[String] @ Dyn via LCMTransport(/words#std_msgs.String)
+\Send/Receive Test:
+<dimos.msgs.std_msgs.String.String object at 0x7f03eb2d10d0>
+```