diff --git a/src/control_backend/agents/actuation/robot_gesture_agent.py b/src/control_backend/agents/actuation/robot_gesture_agent.py
index 997b684..9567940 100644
--- a/src/control_backend/agents/actuation/robot_gesture_agent.py
+++ b/src/control_backend/agents/actuation/robot_gesture_agent.py
@@ -1,4 +1,5 @@
 import json
+import logging
 
 import zmq
 import zmq.asyncio as azmq
@@ -8,6 +9,8 @@ from control_backend.core.agent_system import InternalMessage
 from control_backend.core.config import settings
 from control_backend.schemas.ri_message import GestureCommand, RIEndpoint
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class RobotGestureAgent(BaseAgent):
     """
@@ -111,6 +114,7 @@ class RobotGestureAgent(BaseAgent):
                         gesture_command.data,
                     )
                     return
+            experiment_logger.action("Gesture: %s", gesture_command.data)
             await self.pubsocket.send_json(gesture_command.model_dump())
         except Exception:
             self.logger.exception("Error processing internal message.")
diff --git a/src/control_backend/agents/bdi/bdi_core_agent.py b/src/control_backend/agents/bdi/bdi_core_agent.py
index 54b5149..698bbc4 100644
--- a/src/control_backend/agents/bdi/bdi_core_agent.py
+++ b/src/control_backend/agents/bdi/bdi_core_agent.py
@@ -1,6 +1,7 @@
 import asyncio
 import copy
 import json
+import logging
 import time
 from collections.abc import Iterable
 
@@ -19,6 +20,9 @@ from control_backend.schemas.ri_message import GestureCommand, RIEndpoint, Speec
 DELIMITER = ";\n"  # TODO: temporary until we support lists in AgentSpeak
 
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
+
 class BDICoreAgent(BaseAgent):
     """
     BDI Core Agent.
@@ -207,6 +211,9 @@ class BDICoreAgent(BaseAgent):
         else:
             term = agentspeak.Literal(name)
 
+        if name != "user_said":
+            experiment_logger.observation(f"Formed new belief: {name}{f'={args}' if args else ''}")
+
         self.bdi_agent.call(
             agentspeak.Trigger.addition,
             agentspeak.GoalType.belief,
@@ -244,6 +251,9 @@ class BDICoreAgent(BaseAgent):
             new_args = (agentspeak.Literal(arg) for arg in args)
             term = agentspeak.Literal(name, new_args)
 
+        if name != "user_said":
+            experiment_logger.observation(f"Removed belief: {name}{f'={args}' if args else ''}")
+
         result = self.bdi_agent.call(
             agentspeak.Trigger.removal,
             agentspeak.GoalType.belief,
@@ -386,6 +396,8 @@ class BDICoreAgent(BaseAgent):
                 body=str(message_text),
             )
 
+            experiment_logger.chat(str(message_text), extra={"role": "assistant"})
+
             self.add_behavior(self.send(chat_history_message))
 
             yield
diff --git a/src/control_backend/agents/bdi/bdi_program_manager.py b/src/control_backend/agents/bdi/bdi_program_manager.py
index 6e8a594..3ea6a62 100644
--- a/src/control_backend/agents/bdi/bdi_program_manager.py
+++ b/src/control_backend/agents/bdi/bdi_program_manager.py
@@ -1,10 +1,12 @@
 import asyncio
 import json
+import logging
 
 import zmq
 from pydantic import ValidationError
 from zmq.asyncio import Context
 
+import control_backend
 from control_backend.agents import BaseAgent
 from control_backend.agents.bdi.agentspeak_generator import AgentSpeakGenerator
 from control_backend.core.config import settings
@@ -19,6 +21,8 @@ from control_backend.schemas.program import (
     Program,
 )
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class BDIProgramManager(BaseAgent):
     """
@@ -277,6 +281,18 @@ class BDIProgramManager(BaseAgent):
         await self.send(extractor_msg)
         self.logger.debug("Sent message to extractor agent to clear history.")
 
+    @staticmethod
+    def _rollover_experiment_logs():
+        """
+        A new experiment program started; make a new experiment log file.
+        """
+        handlers = logging.getLogger(settings.logging_settings.experiment_logger_name).handlers
+        for handler in handlers:
+            if isinstance(handler, control_backend.logging.DatedFileHandler):
+                experiment_logger.action("Doing rollover...")
+                handler.do_rollover()
+                experiment_logger.debug("Finished rollover.")
+
     async def _receive_programs(self):
         """
         Continuous loop that receives program updates from the HTTP endpoint.
@@ -297,6 +313,7 @@ class BDIProgramManager(BaseAgent):
             self._initialize_internal_state(program)
             await self._send_program_to_user_interrupt(program)
             await self._send_clear_llm_history()
+            self._rollover_experiment_logs()
 
             await asyncio.gather(
                 self._create_agentspeak_and_send_to_bdi(program),
diff --git a/src/control_backend/agents/llm/llm_agent.py b/src/control_backend/agents/llm/llm_agent.py
index 8d81249..08a77e3 100644
--- a/src/control_backend/agents/llm/llm_agent.py
+++ b/src/control_backend/agents/llm/llm_agent.py
@@ -1,5 +1,6 @@
 import asyncio
 import json
+import logging
 import re
 import uuid
 from collections.abc import AsyncGenerator
@@ -14,6 +15,8 @@ from control_backend.core.config import settings
 from ...schemas.llm_prompt_message import LLMPromptMessage
 from .llm_instructions import LLMInstructions
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class LLMAgent(BaseAgent):
     """
@@ -170,7 +173,7 @@ class LLMAgent(BaseAgent):
             *self.history,
         ]
 
-        message_id = str(uuid.uuid4())  # noqa
+        message_id = str(uuid.uuid4())
 
         try:
             full_message = ""
@@ -179,10 +182,9 @@ class LLMAgent(BaseAgent):
                 full_message += token
                 current_chunk += token
 
-                self.logger.llm(
-                    "Received token: %s",
+                experiment_logger.chat(
                     full_message,
-                    extra={"reference": message_id},  # Used in the UI to update old logs
+                    extra={"role": "assistant", "reference": message_id, "partial": True},
                 )
 
                 # Stream the message in chunks separated by punctuation.
@@ -197,6 +199,11 @@ class LLMAgent(BaseAgent):
             # Yield any remaining tail
             if current_chunk:
                 yield current_chunk
+
+            experiment_logger.chat(
+                full_message,
+                extra={"role": "assistant", "reference": message_id, "partial": False},
+            )
         except httpx.HTTPError as err:
             self.logger.error("HTTP error.", exc_info=err)
             yield "LLM service unavailable."
diff --git a/src/control_backend/agents/perception/transcription_agent/transcription_agent.py b/src/control_backend/agents/perception/transcription_agent/transcription_agent.py
index 795623d..e69fea6 100644
--- a/src/control_backend/agents/perception/transcription_agent/transcription_agent.py
+++ b/src/control_backend/agents/perception/transcription_agent/transcription_agent.py
@@ -1,4 +1,5 @@
 import asyncio
+import logging
 
 import numpy as np
 import zmq
@@ -10,6 +11,8 @@ from control_backend.core.config import settings
 
 from .speech_recognizer import SpeechRecognizer
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class TranscriptionAgent(BaseAgent):
     """
@@ -25,6 +28,8 @@ class TranscriptionAgent(BaseAgent):
     :ivar audio_in_socket: The ZMQ SUB socket instance.
     :ivar speech_recognizer: The speech recognition engine instance.
     :ivar _concurrency: Semaphore to limit concurrent transcriptions.
+    :ivar _current_speech_reference: The reference of the current user utterance, for synchronising
+        experiment logs.
     """
 
     def __init__(self, audio_in_address: str):
@@ -39,6 +44,7 @@ class TranscriptionAgent(BaseAgent):
         self.audio_in_socket: azmq.Socket | None = None
         self.speech_recognizer = None
         self._concurrency = None
+        self._current_speech_reference: str | None = None
 
     async def setup(self):
         """
@@ -63,6 +69,10 @@ class TranscriptionAgent(BaseAgent):
 
         self.logger.info("Finished setting up %s", self.name)
 
+    async def handle_message(self, msg: InternalMessage):
+        if msg.thread == "voice_activity":
+            self._current_speech_reference = msg.body
+
     async def stop(self):
         """
         Stop the agent and close sockets.
@@ -96,24 +106,25 @@ class TranscriptionAgent(BaseAgent):
 
     async def _share_transcription(self, transcription: str):
         """
-        Share a transcription to the other agents that depend on it.
+        Share a transcription to the other agents that depend on it, and to experiment logs.
 
         Currently sends to:
         - :attr:`settings.agent_settings.text_belief_extractor_name`
+        - The UI via the experiment logger
 
         :param transcription: The transcribed text.
         """
-        receiver_names = [
-            settings.agent_settings.text_belief_extractor_name,
-        ]
+        experiment_logger.chat(
+            transcription,
+            extra={"role": "user", "reference": self._current_speech_reference, "partial": False},
+        )
 
-        for receiver_name in receiver_names:
-            message = InternalMessage(
-                to=receiver_name,
-                sender=self.name,
-                body=transcription,
-            )
-            await self.send(message)
+        message = InternalMessage(
+            to=settings.agent_settings.text_belief_extractor_name,
+            sender=self.name,
+            body=transcription,
+        )
+        await self.send(message)
 
     async def _transcribing_loop(self) -> None:
         """
@@ -129,10 +140,9 @@ class TranscriptionAgent(BaseAgent):
                 audio = np.frombuffer(audio_data, dtype=np.float32)
                 speech = await self._transcribe(audio)
                 if not speech:
-                    self.logger.info("Nothing transcribed.")
+                    self.logger.debug("Nothing transcribed.")
                     continue
 
-                self.logger.info("Transcribed speech: %s", speech)
                 await self._share_transcription(speech)
             except Exception as e:
                 self.logger.error(f"Error in transcription loop: {e}")
diff --git a/src/control_backend/agents/perception/vad_agent.py b/src/control_backend/agents/perception/vad_agent.py
index 2b333f5..f397563 100644
--- a/src/control_backend/agents/perception/vad_agent.py
+++ b/src/control_backend/agents/perception/vad_agent.py
@@ -1,4 +1,6 @@
 import asyncio
+import logging
+import uuid
 
 import numpy as np
 import torch
@@ -12,6 +14,8 @@ from control_backend.schemas.internal_message import InternalMessage
 from ...schemas.program_status import PROGRAM_STATUS, ProgramStatus
 from .transcription_agent.transcription_agent import TranscriptionAgent
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class SocketPoller[T]:
     """
@@ -252,6 +256,18 @@ class VADAgent(BaseAgent):
             if prob > prob_threshold:
                 if self.i_since_speech > non_speech_patience + begin_silence_length:
                     self.logger.debug("Speech started.")
+                    reference = str(uuid.uuid4())
+                    experiment_logger.chat(
+                        "...",
+                        extra={"role": "user", "reference": reference, "partial": True},
+                    )
+                    await self.send(
+                        InternalMessage(
+                            to=settings.agent_settings.transcription_name,
+                            body=reference,
+                            thread="voice_activity",
+                        )
+                    )
                 self.audio_buffer = np.append(self.audio_buffer, chunk)
                 self.i_since_speech = 0
                 continue
@@ -269,9 +285,10 @@ class VADAgent(BaseAgent):
                 assert self.audio_out_socket is not None
                 await self.audio_out_socket.send(self.audio_buffer[: -2 * len(chunk)].tobytes())
 
-            # At this point, we know that the speech has ended.
-            # Prepend the last chunk that had no speech, for a more fluent boundary
-            self.audio_buffer = chunk
+            # At this point, we know that there is no speech.
+            # Prepend the last few chunks that had no speech, for a more fluent boundary.
+            self.audio_buffer = np.append(self.audio_buffer, chunk)
+            self.audio_buffer = self.audio_buffer[-begin_silence_length * len(chunk) :]
 
     async def handle_message(self, msg: InternalMessage):
         """
diff --git a/src/control_backend/agents/user_interrupt/user_interrupt_agent.py b/src/control_backend/agents/user_interrupt/user_interrupt_agent.py
index 25f24af..7975aa6 100644
--- a/src/control_backend/agents/user_interrupt/user_interrupt_agent.py
+++ b/src/control_backend/agents/user_interrupt/user_interrupt_agent.py
@@ -1,4 +1,5 @@
 import json
+import logging
 
 import zmq
 from zmq.asyncio import Context
@@ -16,6 +17,8 @@ from control_backend.schemas.ri_message import (
     SpeechCommand,
 )
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class UserInterruptAgent(BaseAgent):
     """
@@ -305,6 +308,7 @@ class UserInterruptAgent(BaseAgent):
 
         :param text_to_say: The string that the robot has to say.
         """
+        experiment_logger.chat(text_to_say, extra={"role": "assistant"})
         cmd = SpeechCommand(data=text_to_say, is_priority=True)
         out_msg = InternalMessage(
             to=settings.agent_settings.robot_speech_name,
diff --git a/test/unit/agents/actuation/test_robot_gesture_agent.py b/test/unit/agents/actuation/test_robot_gesture_agent.py
index 1e6fd8a..20d7d51 100644
--- a/test/unit/agents/actuation/test_robot_gesture_agent.py
+++ b/test/unit/agents/actuation/test_robot_gesture_agent.py
@@ -1,5 +1,5 @@
 import json
-from unittest.mock import AsyncMock, MagicMock
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 import zmq
@@ -19,6 +19,12 @@ def zmq_context(mocker):
     return mock_context
 
 
+@pytest.fixture(autouse=True)
+def mock_experiment_logger():
+    with patch("control_backend.agents.actuation.robot_gesture_agent.experiment_logger") as logger:
+        yield logger
+
+
 @pytest.mark.asyncio
 async def test_setup_bind(zmq_context, mocker):
     """Setup binds and subscribes to internal commands."""
diff --git a/test/unit/agents/bdi/test_bdi_core_agent.py b/test/unit/agents/bdi/test_bdi_core_agent.py
index 6245d5b..1bf0107 100644
--- a/test/unit/agents/bdi/test_bdi_core_agent.py
+++ b/test/unit/agents/bdi/test_bdi_core_agent.py
@@ -26,6 +26,12 @@ def agent():
     return agent
 
 
+@pytest.fixture(autouse=True)
+def mock_experiment_logger():
+    with patch("control_backend.agents.bdi.bdi_core_agent.experiment_logger") as logger:
+        yield logger
+
+
 @pytest.mark.asyncio
 async def test_setup_loads_asl(mock_agentspeak_env, agent):
     # Mock file opening
diff --git a/test/unit/agents/llm/test_llm_agent.py b/test/unit/agents/llm/test_llm_agent.py
index bd407cc..bbd6e93 100644
--- a/test/unit/agents/llm/test_llm_agent.py
+++ b/test/unit/agents/llm/test_llm_agent.py
@@ -18,6 +18,12 @@ def mock_httpx_client():
         yield mock_client
 
 
+@pytest.fixture(autouse=True)
+def mock_experiment_logger():
+    with patch("control_backend.agents.llm.llm_agent.experiment_logger") as logger:
+        yield logger
+
+
 @pytest.mark.asyncio
 async def test_llm_processing_success(mock_httpx_client, mock_settings):
     # Setup the mock response for the stream
diff --git a/test/unit/agents/perception/transcription_agent/test_transcription_agent.py b/test/unit/agents/perception/transcription_agent/test_transcription_agent.py
index 57875ca..f5a4d1c 100644
--- a/test/unit/agents/perception/transcription_agent/test_transcription_agent.py
+++ b/test/unit/agents/perception/transcription_agent/test_transcription_agent.py
@@ -14,6 +14,15 @@ from control_backend.agents.perception.transcription_agent.transcription_agent i
 )
 
 
+@pytest.fixture(autouse=True)
+def mock_experiment_logger():
+    with patch(
+        "control_backend.agents.perception"
+        ".transcription_agent.transcription_agent.experiment_logger"
+    ) as logger:
+        yield logger
+
+
 @pytest.mark.asyncio
 async def test_transcription_agent_flow(mock_zmq_context):
     mock_sub = MagicMock()
diff --git a/test/unit/agents/perception/vad_agent/test_vad_streaming.py b/test/unit/agents/perception/vad_agent/test_vad_streaming.py
index 349fab2..b53f63d 100644
--- a/test/unit/agents/perception/vad_agent/test_vad_streaming.py
+++ b/test/unit/agents/perception/vad_agent/test_vad_streaming.py
@@ -24,7 +24,9 @@ def audio_out_socket():
 
 @pytest.fixture
 def vad_agent(audio_out_socket):
-    return VADAgent("tcp://localhost:5555", False)
+    agent = VADAgent("tcp://localhost:5555", False)
+    agent._internal_pub_socket = AsyncMock()
+    return agent
 
 
 @pytest.fixture(autouse=True)
@@ -44,6 +46,12 @@ def patch_settings(monkeypatch):
     monkeypatch.setattr(vad_agent.settings.vad_settings, "sample_rate_hz", 16_000, raising=False)
 
 
+@pytest.fixture(autouse=True)
+def mock_experiment_logger():
+    with patch("control_backend.agents.perception.vad_agent.experiment_logger") as logger:
+        yield logger
+
+
 async def simulate_streaming_with_probabilities(streaming, probabilities: list[float]):
     """
     Simulates a streaming scenario with given VAD model probabilities for testing purposes.
@@ -84,14 +92,15 @@ async def test_voice_activity_detected(audio_out_socket, vad_agent):
     Test a scenario where there is voice activity detected between silences.
     """
     speech_chunk_count = 5
-    probabilities = [0.0] * 5 + [1.0] * speech_chunk_count + [0.0] * 5
+    begin_silence_chunks = settings.behaviour_settings.vad_begin_silence_chunks
+    probabilities = [0.0] * 15 + [1.0] * speech_chunk_count + [0.0] * 5
     vad_agent.audio_out_socket = audio_out_socket
     await simulate_streaming_with_probabilities(vad_agent, probabilities)
 
     audio_out_socket.send.assert_called_once()
     data = audio_out_socket.send.call_args[0][0]
     assert isinstance(data, bytes)
-    assert len(data) == 512 * 4 * (speech_chunk_count + 1)
+    assert len(data) == 512 * 4 * (begin_silence_chunks + speech_chunk_count)
 
 
 @pytest.mark.asyncio
@@ -101,8 +110,9 @@ async def test_voice_activity_short_pause(audio_out_socket, vad_agent):
     short pause.
     """
     speech_chunk_count = 5
+    begin_silence_chunks = settings.behaviour_settings.vad_begin_silence_chunks
     probabilities = (
-        [0.0] * 5 + [1.0] * speech_chunk_count + [0.0] + [1.0] * speech_chunk_count + [0.0] * 5
+        [0.0] * 15 + [1.0] * speech_chunk_count + [0.0] + [1.0] * speech_chunk_count + [0.0] * 5
     )
     vad_agent.audio_out_socket = audio_out_socket
     await simulate_streaming_with_probabilities(vad_agent, probabilities)
@@ -110,8 +120,8 @@ async def test_voice_activity_short_pause(audio_out_socket, vad_agent):
     audio_out_socket.send.assert_called_once()
     data = audio_out_socket.send.call_args[0][0]
     assert isinstance(data, bytes)
-    # Expecting 13 chunks (2*5 with speech, 1 pause between, 1 as padding)
-    assert len(data) == 512 * 4 * (speech_chunk_count * 2 + 1 + 1)
+    # Expecting 13 chunks (2*5 with speech, 1 pause between, begin_silence_chunks as padding)
+    assert len(data) == 512 * 4 * (speech_chunk_count * 2 + 1 + begin_silence_chunks)
 
 
 @pytest.mark.asyncio
diff --git a/test/unit/agents/user_interrupt/test_user_interrupt.py b/test/unit/agents/user_interrupt/test_user_interrupt.py
index 3786f8d..c41d79e 100644
--- a/test/unit/agents/user_interrupt/test_user_interrupt.py
+++ b/test/unit/agents/user_interrupt/test_user_interrupt.py
@@ -30,6 +30,14 @@ def agent():
     return agent
 
 
+@pytest.fixture(autouse=True)
+def mock_experiment_logger():
+    with patch(
+        "control_backend.agents.user_interrupt.user_interrupt_agent.experiment_logger"
+    ) as logger:
+        yield logger
+
+
 @pytest.mark.asyncio
 async def test_send_to_speech_agent(agent):
     """Verify speech command format."""
@@ -579,30 +587,6 @@ async def test_create_mapping_recursive_goals(agent):
     assert agent._goal_reverse_map["child_goal"] == str(child_goal_id)
 
 
-@pytest.mark.asyncio
-async def test_setup(agent):
-    """Test the setup method initializes sockets correctly."""
-    with patch("control_backend.agents.user_interrupt.user_interrupt_agent.Context") as MockContext:
-        mock_ctx_instance = MagicMock()
-        MockContext.instance.return_value = mock_ctx_instance
-
-        mock_sub = MagicMock()
-        mock_pub = MagicMock()
-        mock_ctx_instance.socket.side_effect = [mock_sub, mock_pub]
-
-        # MOCK add_behavior so we don't rely on internal attributes
-        agent.add_behavior = MagicMock()
-
-        await agent.setup()
-
-        # Check sockets
-        mock_sub.connect.assert_called_with(settings.zmq_settings.internal_sub_address)
-        mock_pub.connect.assert_called_with(settings.zmq_settings.internal_pub_address)
-
-        # Verify add_behavior was called
-        agent.add_behavior.assert_called_once()
-
-
 @pytest.mark.asyncio
 async def test_receive_loop_advanced_scenarios(agent):
     """
@@ -706,63 +690,3 @@ async def test_receive_loop_advanced_scenarios(agent):
 
     # Next Phase
     agent._send_experiment_control_to_bdi_core.assert_awaited_with("next_phase")
-
-
-@pytest.mark.asyncio
-async def test_handle_message_unknown_thread(agent):
-    """Test handling of an unknown message thread (lines 213-214)."""
-    msg = InternalMessage(to="me", thread="unknown_thread", body="test")
-    await agent.handle_message(msg)
-
-    agent.logger.debug.assert_called_with(
-        "Received internal message on unhandled thread: unknown_thread"
-    )
-
-
-@pytest.mark.asyncio
-async def test_send_to_bdi_belief_edge_cases(agent):
-    """
-    Covers:
-    - Unknown asl_type warning (lines 326-328)
-    - unachieve=True logic (lines 334-337)
-    """
-    # 1. Unknown Type
-    await agent._send_to_bdi_belief("slug", "unknown_type")
-
-    agent.logger.warning.assert_called_with("Tried to send belief with unknown type")
-    agent.send.assert_not_called()
-
-    # Reset mock for part 2
-    agent.send.reset_mock()
-
-    # 2. Unachieve = True
-    await agent._send_to_bdi_belief("slug", "cond_norm", unachieve=True)
-
-    agent.send.assert_awaited()
-    sent_msg = agent.send.call_args.args[0]
-
-    # Verify it is a delete operation
-    body_obj = BeliefMessage.model_validate_json(sent_msg.body)
-
-    # Verify 'delete' has content
-    assert body_obj.delete is not None
-    assert len(body_obj.delete) == 1
-    assert body_obj.delete[0].name == "force_slug"
-
-    # Verify 'create' is empty (handling both None and [])
-    assert not body_obj.create
-
-
-@pytest.mark.asyncio
-async def test_send_experiment_control_unknown(agent):
-    """Test sending an unknown experiment control type (lines 366-367)."""
-    await agent._send_experiment_control_to_bdi_core("invalid_command")
-
-    agent.logger.warning.assert_called_with(
-        "Received unknown experiment control type '%s' to send to BDI Core.", "invalid_command"
-    )
-
-    # Ensure it still sends an empty message (as per code logic, though thread is empty)
-    agent.send.assert_awaited()
-    msg = agent.send.call_args[0][0]
-    assert msg.thread == ""