From 9cc0e3995585274828759ecc6870ce4b4c234e17 Mon Sep 17 00:00:00 2001 From: Twirre Meulenbelt <43213592+TwirreM@users.noreply.github.com> Date: Thu, 11 Dec 2025 12:04:24 +0100 Subject: [PATCH 1/5] fix: failures main tests since VAD agent initialization was changed The test still expects the VAD agent to be started in main, rather than in the RI Communication Agent. ref: N25B-356 --- test/unit/test_main.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/test/unit/test_main.py b/test/unit/test_main.py index 2737c76..a423703 100644 --- a/test/unit/test_main.py +++ b/test/unit/test_main.py @@ -53,8 +53,6 @@ async def test_lifespan_agent_start_exception(): Ensures exceptions are logged properly and re-raised. """ with ( - patch("control_backend.main.VADAgent.start", new_callable=AsyncMock), - patch("control_backend.main.VADAgent.reset_stream", new_callable=AsyncMock), patch( "control_backend.main.RICommunicationAgent.start", new_callable=AsyncMock ) as ri_start, From 32d8f20dc964cc416b81b00af94b242916876e88 Mon Sep 17 00:00:00 2001 From: Twirre Meulenbelt <43213592+TwirreM@users.noreply.github.com> Date: Thu, 11 Dec 2025 12:12:15 +0100 Subject: [PATCH 2/5] feat: parameterize RI host Was "localhost" in RI Communication Agent, now uses configurable setting. Secretly also removing "localhost" from VAD agent, as its socket should be something that's "inproc". ref: N25B-352 --- .../communication/ri_communication_agent.py | 2 +- .../agents/perception/vad_agent.py | 13 +++++++------ src/control_backend/core/config.py | 3 +++ .../agents/perception/vad_agent/test_vad_agent.py | 2 +- .../perception/vad_agent/test_vad_streaming.py | 15 +++++++++++---- 5 files changed, 23 insertions(+), 12 deletions(-) diff --git a/src/control_backend/agents/communication/ri_communication_agent.py b/src/control_backend/agents/communication/ri_communication_agent.py index 9f4bc4d..3bca6e4 100644 --- a/src/control_backend/agents/communication/ri_communication_agent.py +++ b/src/control_backend/agents/communication/ri_communication_agent.py @@ -167,7 +167,7 @@ class RICommunicationAgent(BaseAgent): bind = port_data["bind"] if not bind: - addr = f"tcp://localhost:{port}" + addr = f"tcp://{settings.ri_host}:{port}" else: addr = f"tcp://*:{port}" diff --git a/src/control_backend/agents/perception/vad_agent.py b/src/control_backend/agents/perception/vad_agent.py index 8ccff0a..5d0c497 100644 --- a/src/control_backend/agents/perception/vad_agent.py +++ b/src/control_backend/agents/perception/vad_agent.py @@ -103,12 +103,11 @@ class VADAgent(BaseAgent): self._connect_audio_in_socket() - audio_out_port = self._connect_audio_out_socket() - if audio_out_port is None: + audio_out_address = self._connect_audio_out_socket() + if audio_out_address is None: self.logger.error("Could not bind output socket, stopping.") await self.stop() return - audio_out_address = f"tcp://localhost:{audio_out_port}" # Connect to internal communication socket self.program_sub_socket = azmq.Context.instance().socket(zmq.SUB) @@ -161,13 +160,15 @@ class VADAgent(BaseAgent): self.audio_in_socket.connect(self.audio_in_address) self.audio_in_poller = SocketPoller[bytes](self.audio_in_socket) - def _connect_audio_out_socket(self) -> int | None: + def _connect_audio_out_socket(self) -> str | None: """ - Returns the port bound, or None if binding failed. + Returns the address that was bound to, or None if binding failed. """ try: self.audio_out_socket = azmq.Context.instance().socket(zmq.PUB) - return self.audio_out_socket.bind_to_random_port("tcp://localhost", max_tries=100) + address = "inproc://vad_stream" + self.audio_out_socket.bind(address) + return address except zmq.ZMQBindError: self.logger.error("Failed to bind an audio output socket after 100 tries.") self.audio_out_socket = None diff --git a/src/control_backend/core/config.py b/src/control_backend/core/config.py index fa105a5..0154c28 100644 --- a/src/control_backend/core/config.py +++ b/src/control_backend/core/config.py @@ -125,6 +125,7 @@ class Settings(BaseSettings): :ivar app_title: Title of the application. :ivar ui_url: URL of the frontend UI. + :ivar ui_url: The hostname of the Robot Interface. :ivar zmq_settings: ZMQ configuration. :ivar agent_settings: Agent name configuration. :ivar behaviour_settings: Behavior configuration. @@ -137,6 +138,8 @@ class Settings(BaseSettings): ui_url: str = "http://localhost:5173" + ri_host: str = "localhost" + zmq_settings: ZMQSettings = ZMQSettings() agent_settings: AgentSettings = AgentSettings() diff --git a/test/integration/agents/perception/vad_agent/test_vad_agent.py b/test/integration/agents/perception/vad_agent/test_vad_agent.py index f5f2615..668d1ce 100644 --- a/test/integration/agents/perception/vad_agent/test_vad_agent.py +++ b/test/integration/agents/perception/vad_agent/test_vad_agent.py @@ -91,7 +91,7 @@ def test_out_socket_creation(zmq_context): assert per_vad_agent.audio_out_socket is not None zmq_context.return_value.socket.assert_called_once_with(zmq.PUB) - zmq_context.return_value.socket.return_value.bind_to_random_port.assert_called_once() + zmq_context.return_value.socket.return_value.bind.assert_called_once_with("inproc://vad_stream") @pytest.mark.asyncio 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 4440cae..166919f 100644 --- a/test/unit/agents/perception/vad_agent/test_vad_streaming.py +++ b/test/unit/agents/perception/vad_agent/test_vad_streaming.py @@ -7,6 +7,15 @@ import zmq from control_backend.agents.perception.vad_agent import VADAgent +# We don't want to use real ZMQ in unit tests, for example because it can give errors when sockets +# aren't closed properly. +@pytest.fixture(autouse=True) +def mock_zmq(): + with patch("zmq.asyncio.Context") as mock: + mock.instance.return_value = MagicMock() + yield mock + + @pytest.fixture def audio_out_socket(): return AsyncMock() @@ -140,12 +149,10 @@ async def test_vad_model_load_failure_stops_agent(vad_agent): # Patch stop to an AsyncMock so we can check it was awaited vad_agent.stop = AsyncMock() - result = await vad_agent.setup() + await vad_agent.setup() # Assert stop was called vad_agent.stop.assert_awaited_once() - # Assert setup returned None - assert result is None @pytest.mark.asyncio @@ -155,7 +162,7 @@ async def test_audio_out_bind_failure_sets_none_and_logs(vad_agent, caplog): audio_out_socket is set to None, None is returned, and an error is logged. """ mock_socket = MagicMock() - mock_socket.bind_to_random_port.side_effect = zmq.ZMQBindError() + mock_socket.bind.side_effect = zmq.ZMQBindError() with patch("control_backend.agents.perception.vad_agent.azmq.Context.instance") as mock_ctx: mock_ctx.return_value.socket.return_value = mock_socket From 0c682d6440485007ed0aed885627d8d16a4750f9 Mon Sep 17 00:00:00 2001 From: Twirre Meulenbelt <43213592+TwirreM@users.noreply.github.com> Date: Thu, 11 Dec 2025 13:35:19 +0100 Subject: [PATCH 3/5] feat: introduce .env.example, docs The example includes options that are expected to be changed. It also includes a reference to where in the docs you can find a full list of options. ref: N25B-352 --- .env.example | 20 +++++++++++++ README.md | 9 ++++++ .../agents/actuation/robot_speech_agent.py | 2 +- .../communication/ri_communication_agent.py | 2 +- .../agents/perception/vad_agent.py | 5 ++-- src/control_backend/core/config.py | 30 +++++++++++++++---- .../actuation/test_robot_speech_agent.py | 10 +++---- 7 files changed, 63 insertions(+), 15 deletions(-) create mode 100644 .env.example diff --git a/.env.example b/.env.example new file mode 100644 index 0000000..d498054 --- /dev/null +++ b/.env.example @@ -0,0 +1,20 @@ +# Example .env file. To use, make a copy, call it ".env" (i.e. removing the ".example" suffix), then you edit values. + +# The hostname of the Robot Interface. Change if the Control Backend and Robot Interface are running on different computers. +RI_HOST="localhost" + +# URL for the local LLM API. Must be an API that implements the OpenAI Chat Completions API, but most do. +LLM_SETTINGS__LOCAL_LLM_URL="http://localhost:1234/v1/chat/completions" + +# Name of the local LLM model to use. +LLM_SETTINGS__LOCAL_LLM_MODEL="gpt-oss" + +# Number of non-speech chunks to wait before speech ended. A chunk is approximately 31 ms. Increasing this number allows longer pauses in speech, but also increases response time. +BEHAVIOUR_SETTINGS__VAD_NON_SPEECH_PATIENCE_CHUNKS=3 + +# Timeout in milliseconds for socket polling. Increase this number if network latency/jitter is high, often the case when using Wi-Fi. Perhaps 500 ms. A symptom of this issue is transcriptions getting cut off. +BEHAVIOUR_SETTINGS__SOCKET_POLLER_TIMEOUT_MS=100 + + + +# For an exhaustive list of options, see the control_backend.core.config module in the docs. diff --git a/README.md b/README.md index 1527215..03dac9a 100644 --- a/README.md +++ b/README.md @@ -27,6 +27,7 @@ This + part might differ based on what model you choose. copy the model name in the module loaded and replace local_llm_modelL. In settings. + ## Running To run the project (development server), execute the following command (while inside the root repository): @@ -34,6 +35,14 @@ To run the project (development server), execute the following command (while in uv run fastapi dev src/control_backend/main.py ``` +### Environment Variables + +You can use environment variables to change settings. Make a copy of the [`.env.example`](.env.example) file, name it `.env` and put it in the root directory. The file itself describes how to do the configuration. + +For an exhaustive list of environment options, see the `control_backend.core.config` module in the docs. + + + ## Testing Testing happens automatically when opening a merge request to any branch. If you want to manually run the test suite, you can do so by running the following for unit tests: diff --git a/src/control_backend/agents/actuation/robot_speech_agent.py b/src/control_backend/agents/actuation/robot_speech_agent.py index 674b270..f8e3d4c 100644 --- a/src/control_backend/agents/actuation/robot_speech_agent.py +++ b/src/control_backend/agents/actuation/robot_speech_agent.py @@ -29,7 +29,7 @@ class RobotSpeechAgent(BaseAgent): def __init__( self, name: str, - address=settings.zmq_settings.ri_command_address, + address: str, bind=False, ): super().__init__(name) diff --git a/src/control_backend/agents/communication/ri_communication_agent.py b/src/control_backend/agents/communication/ri_communication_agent.py index 3bca6e4..4a043e9 100644 --- a/src/control_backend/agents/communication/ri_communication_agent.py +++ b/src/control_backend/agents/communication/ri_communication_agent.py @@ -37,7 +37,7 @@ class RICommunicationAgent(BaseAgent): def __init__( self, name: str, - address=settings.zmq_settings.ri_command_address, + address=settings.zmq_settings.ri_communication_address, bind=False, ): super().__init__(name) diff --git a/src/control_backend/agents/perception/vad_agent.py b/src/control_backend/agents/perception/vad_agent.py index 5d0c497..70fa9e1 100644 --- a/src/control_backend/agents/perception/vad_agent.py +++ b/src/control_backend/agents/perception/vad_agent.py @@ -166,9 +166,8 @@ class VADAgent(BaseAgent): """ try: self.audio_out_socket = azmq.Context.instance().socket(zmq.PUB) - address = "inproc://vad_stream" - self.audio_out_socket.bind(address) - return address + self.audio_out_socket.bind(settings.zmq_settings.vad_pub_address) + return settings.zmq_settings.vad_pub_address except zmq.ZMQBindError: self.logger.error("Failed to bind an audio output socket after 100 tries.") self.audio_out_socket = None diff --git a/src/control_backend/core/config.py b/src/control_backend/core/config.py index 0154c28..35acf96 100644 --- a/src/control_backend/core/config.py +++ b/src/control_backend/core/config.py @@ -1,3 +1,12 @@ +""" +An exhaustive overview of configurable options. All of these can be set using environment variables +by nesting with double underscores (__). Start from the ``Settings`` class. + +For example, ``settings.ri_host`` becomes ``RI_HOST``, and +``settings.zmq_settings.ri_communication_address`` becomes +``ZMQ_SETTINGS__RI_COMMUNICATION_ADDRESS``. +""" + from pydantic import BaseModel from pydantic_settings import BaseSettings, SettingsConfigDict @@ -8,15 +17,16 @@ class ZMQSettings(BaseModel): :ivar internal_pub_address: Address for the internal PUB socket. :ivar internal_sub_address: Address for the internal SUB socket. - :ivar ri_command_address: Address for sending commands to the Robot Interface. - :ivar ri_communication_address: Address for receiving communication from the Robot Interface. - :ivar vad_agent_address: Address for the Voice Activity Detection (VAD) agent. + :ivar ri_communication_address: Address for the endpoint that the Robot Interface connects to. + :ivar vad_pub_address: Address that the VAD agent binds to and publishes audio segments to. """ + # ATTENTION: When adding/removing settings, make sure to update the .env.example file + internal_pub_address: str = "tcp://localhost:5560" internal_sub_address: str = "tcp://localhost:5561" - ri_command_address: str = "tcp://localhost:0000" ri_communication_address: str = "tcp://*:5555" + vad_pub_address: str = "inproc://vad_stream" class AgentSettings(BaseModel): @@ -35,6 +45,8 @@ class AgentSettings(BaseModel): :ivar robot_speech_name: Name of the Robot Speech Agent. """ + # ATTENTION: When adding/removing settings, make sure to update the .env.example file + # agent names bdi_core_name: str = "bdi_core_agent" bdi_belief_collector_name: str = "belief_collector_agent" @@ -64,6 +76,8 @@ class BehaviourSettings(BaseModel): :ivar transcription_token_buffer: Buffer for transcription tokens. """ + # ATTENTION: When adding/removing settings, make sure to update the .env.example file + sleep_s: float = 1.0 comm_setup_max_retries: int = 5 socket_poller_timeout_ms: int = 100 @@ -88,6 +102,8 @@ class LLMSettings(BaseModel): :ivar local_llm_model: Name of the local LLM model to use. """ + # ATTENTION: When adding/removing settings, make sure to update the .env.example file + local_llm_url: str = "http://localhost:1234/v1/chat/completions" local_llm_model: str = "gpt-oss" @@ -101,6 +117,8 @@ class VADSettings(BaseModel): :ivar sample_rate_hz: Sample rate in Hz for the VAD model. """ + # ATTENTION: When adding/removing settings, make sure to update the .env.example file + repo_or_dir: str = "snakers4/silero-vad" model_name: str = "silero_vad" sample_rate_hz: int = 16000 @@ -114,6 +132,8 @@ class SpeechModelSettings(BaseModel): :ivar openai_model_name: Model name for OpenAI-based speech recognition. """ + # ATTENTION: When adding/removing settings, make sure to update the .env.example file + # model identifiers for speech recognition mlx_model_name: str = "mlx-community/whisper-small.en-mlx" openai_model_name: str = "small.en" @@ -125,7 +145,7 @@ class Settings(BaseSettings): :ivar app_title: Title of the application. :ivar ui_url: URL of the frontend UI. - :ivar ui_url: The hostname of the Robot Interface. + :ivar ri_host: The hostname of the Robot Interface. :ivar zmq_settings: ZMQ configuration. :ivar agent_settings: Agent name configuration. :ivar behaviour_settings: Behavior configuration. diff --git a/test/unit/agents/actuation/test_robot_speech_agent.py b/test/unit/agents/actuation/test_robot_speech_agent.py index 15324f6..567a5a1 100644 --- a/test/unit/agents/actuation/test_robot_speech_agent.py +++ b/test/unit/agents/actuation/test_robot_speech_agent.py @@ -56,7 +56,7 @@ async def test_setup_connect(zmq_context, mocker): async def test_handle_message_sends_command(): """Internal message is forwarded to robot pub socket as JSON.""" pubsocket = AsyncMock() - agent = RobotSpeechAgent("robot_speech") + agent = RobotSpeechAgent("robot_speech", "tcp://localhost:3498") agent.pubsocket = pubsocket payload = {"endpoint": "actuate/speech", "data": "hello"} @@ -80,7 +80,7 @@ async def test_zmq_command_loop_valid_payload(zmq_context): fake_socket.recv_multipart = recv_once fake_socket.send_json = AsyncMock() - agent = RobotSpeechAgent("robot_speech") + agent = RobotSpeechAgent("robot_speech", "tcp://localhost:3498") agent.subsocket = fake_socket agent.pubsocket = fake_socket agent._running = True @@ -101,7 +101,7 @@ async def test_zmq_command_loop_invalid_json(): fake_socket.recv_multipart = recv_once fake_socket.send_json = AsyncMock() - agent = RobotSpeechAgent("robot_speech") + agent = RobotSpeechAgent("robot_speech", "tcp://localhost:3498") agent.subsocket = fake_socket agent.pubsocket = fake_socket agent._running = True @@ -115,7 +115,7 @@ async def test_zmq_command_loop_invalid_json(): async def test_handle_message_invalid_payload(): """Invalid payload is caught and does not send.""" pubsocket = AsyncMock() - agent = RobotSpeechAgent("robot_speech") + agent = RobotSpeechAgent("robot_speech", "tcp://localhost:3498") agent.pubsocket = pubsocket msg = InternalMessage(to="robot", sender="tester", body=json.dumps({"bad": "data"})) @@ -129,7 +129,7 @@ async def test_handle_message_invalid_payload(): async def test_stop_closes_sockets(): pubsocket = MagicMock() subsocket = MagicMock() - agent = RobotSpeechAgent("robot_speech") + agent = RobotSpeechAgent("robot_speech", "tcp://localhost:3498") agent.pubsocket = pubsocket agent.subsocket = subsocket From 7d798f2e77da8c862a6e9aba91c380e22a14bd86 Mon Sep 17 00:00:00 2001 From: Twirre Meulenbelt <43213592+TwirreM@users.noreply.github.com> Date: Mon, 29 Dec 2025 12:40:16 +0100 Subject: [PATCH 4/5] Merge remote-tracking branch 'origin/dev' into feat/environment-variables # Conflicts: # src/control_backend/core/config.py # test/unit/agents/actuation/test_robot_speech_agent.py --- .../agents/actuation/robot_gesture_agent.py | 2 +- .../actuation/test_robot_gesture_agent.py | 50 ++++++++++--------- 2 files changed, 28 insertions(+), 24 deletions(-) diff --git a/src/control_backend/agents/actuation/robot_gesture_agent.py b/src/control_backend/agents/actuation/robot_gesture_agent.py index 4f5dd79..3b264d2 100644 --- a/src/control_backend/agents/actuation/robot_gesture_agent.py +++ b/src/control_backend/agents/actuation/robot_gesture_agent.py @@ -33,7 +33,7 @@ class RobotGestureAgent(BaseAgent): def __init__( self, name: str, - address=settings.zmq_settings.ri_command_address, + address: str, bind=False, gesture_data=None, single_gesture_data=None, diff --git a/test/unit/agents/actuation/test_robot_gesture_agent.py b/test/unit/agents/actuation/test_robot_gesture_agent.py index c68f052..fe051a6 100644 --- a/test/unit/agents/actuation/test_robot_gesture_agent.py +++ b/test/unit/agents/actuation/test_robot_gesture_agent.py @@ -73,7 +73,7 @@ async def test_setup_connect(zmq_context, mocker): async def test_handle_message_sends_valid_gesture_command(): """Internal message with valid gesture tag is forwarded to robot pub socket.""" pubsocket = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.pubsocket = pubsocket payload = { @@ -91,7 +91,7 @@ async def test_handle_message_sends_valid_gesture_command(): async def test_handle_message_sends_non_gesture_command(): """Internal message with non-gesture endpoint is not forwarded by this agent.""" pubsocket = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.pubsocket = pubsocket payload = {"endpoint": "some_other_endpoint", "data": "invalid_tag_not_in_list"} @@ -107,7 +107,7 @@ async def test_handle_message_sends_non_gesture_command(): async def test_handle_message_rejects_invalid_gesture_tag(): """Internal message with invalid gesture tag is not forwarded.""" pubsocket = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.pubsocket = pubsocket # Use a tag that's not in gesture_data @@ -123,7 +123,7 @@ async def test_handle_message_rejects_invalid_gesture_tag(): async def test_handle_message_invalid_payload(): """Invalid payload is caught and does not send.""" pubsocket = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.pubsocket = pubsocket msg = InternalMessage(to="robot", sender="tester", body=json.dumps({"bad": "data"})) @@ -142,12 +142,12 @@ async def test_zmq_command_loop_valid_gesture_payload(): async def recv_once(): # stop after first iteration agent._running = False - return (b"command", json.dumps(command).encode("utf-8")) + return b"command", json.dumps(command).encode("utf-8") fake_socket.recv_multipart = recv_once fake_socket.send_json = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.subsocket = fake_socket agent.pubsocket = fake_socket agent._running = True @@ -165,12 +165,12 @@ async def test_zmq_command_loop_valid_non_gesture_payload(): async def recv_once(): agent._running = False - return (b"command", json.dumps(command).encode("utf-8")) + return b"command", json.dumps(command).encode("utf-8") fake_socket.recv_multipart = recv_once fake_socket.send_json = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.subsocket = fake_socket agent.pubsocket = fake_socket agent._running = True @@ -188,12 +188,12 @@ async def test_zmq_command_loop_invalid_gesture_tag(): async def recv_once(): agent._running = False - return (b"command", json.dumps(command).encode("utf-8")) + return b"command", json.dumps(command).encode("utf-8") fake_socket.recv_multipart = recv_once fake_socket.send_json = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.subsocket = fake_socket agent.pubsocket = fake_socket agent._running = True @@ -210,12 +210,12 @@ async def test_zmq_command_loop_invalid_json(): async def recv_once(): agent._running = False - return (b"command", b"{not_json}") + return b"command", b"{not_json}" fake_socket.recv_multipart = recv_once fake_socket.send_json = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.subsocket = fake_socket agent.pubsocket = fake_socket agent._running = True @@ -232,12 +232,12 @@ async def test_zmq_command_loop_ignores_send_gestures_topic(): async def recv_once(): agent._running = False - return (b"send_gestures", b"{}") + return b"send_gestures", b"{}" fake_socket.recv_multipart = recv_once fake_socket.send_json = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.subsocket = fake_socket agent.pubsocket = fake_socket agent._running = True @@ -259,7 +259,9 @@ async def test_fetch_gestures_loop_without_amount(): fake_repsocket.recv = recv_once fake_repsocket.send = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no", "wave", "point"]) + agent = RobotGestureAgent( + "robot_gesture", gesture_data=["hello", "yes", "no", "wave", "point"], address="" + ) agent.repsocket = fake_repsocket agent._running = True @@ -287,7 +289,9 @@ async def test_fetch_gestures_loop_with_amount(): fake_repsocket.recv = recv_once fake_repsocket.send = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no", "wave", "point"]) + agent = RobotGestureAgent( + "robot_gesture", gesture_data=["hello", "yes", "no", "wave", "point"], address="" + ) agent.repsocket = fake_repsocket agent._running = True @@ -315,7 +319,7 @@ async def test_fetch_gestures_loop_with_integer_request(): fake_repsocket.recv = recv_once fake_repsocket.send = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.repsocket = fake_repsocket agent._running = True @@ -340,7 +344,7 @@ async def test_fetch_gestures_loop_with_invalid_json(): fake_repsocket.recv = recv_once fake_repsocket.send = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.repsocket = fake_repsocket agent._running = True @@ -365,7 +369,7 @@ async def test_fetch_gestures_loop_with_non_integer_json(): fake_repsocket.recv = recv_once fake_repsocket.send = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.repsocket = fake_repsocket agent._running = True @@ -381,7 +385,7 @@ async def test_fetch_gestures_loop_with_non_integer_json(): def test_gesture_data_attribute(): """Test that gesture_data returns the expected list.""" gesture_data = ["hello", "yes", "no", "wave"] - agent = RobotGestureAgent("robot_gesture", gesture_data=gesture_data) + agent = RobotGestureAgent("robot_gesture", gesture_data=gesture_data, address="") assert agent.gesture_data == gesture_data assert isinstance(agent.gesture_data, list) @@ -398,7 +402,7 @@ async def test_stop_closes_sockets(): pubsocket = MagicMock() subsocket = MagicMock() repsocket = MagicMock() - agent = RobotGestureAgent("robot_gesture") + agent = RobotGestureAgent("robot_gesture", address="") agent.pubsocket = pubsocket agent.subsocket = subsocket agent.repsocket = repsocket @@ -415,7 +419,7 @@ async def test_stop_closes_sockets(): async def test_initialization_with_custom_gesture_data(): """Agent can be initialized with custom gesture data.""" custom_gestures = ["custom1", "custom2", "custom3"] - agent = RobotGestureAgent("robot_gesture", gesture_data=custom_gestures) + agent = RobotGestureAgent("robot_gesture", gesture_data=custom_gestures, address="") assert agent.gesture_data == custom_gestures @@ -432,7 +436,7 @@ async def test_fetch_gestures_loop_handles_exception(): fake_repsocket.recv = recv_once fake_repsocket.send = AsyncMock() - agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"]) + agent = RobotGestureAgent("robot_gesture", gesture_data=["hello", "yes", "no"], address="") agent.repsocket = fake_repsocket agent.logger = MagicMock() agent._running = True From 6ca86e4b819c706cd15b7ef7fb06af0b26100019 Mon Sep 17 00:00:00 2001 From: Pim Hutting Date: Fri, 2 Jan 2026 15:13:04 +0000 Subject: [PATCH 5/5] feat: made program reset LLM --- .../agents/bdi/bdi_program_manager.py | 21 ++++++++++++++++-- src/control_backend/agents/llm/llm_agent.py | 4 ++++ .../agents/bdi/test_bdi_program_manager.py | 22 +++++++++++++++++++ test/unit/agents/llm/test_llm_agent.py | 20 +++++++++++++++++ test/unit/test_main.py | 2 -- 5 files changed, 65 insertions(+), 4 deletions(-) diff --git a/src/control_backend/agents/bdi/bdi_program_manager.py b/src/control_backend/agents/bdi/bdi_program_manager.py index 83dea93..2f4f850 100644 --- a/src/control_backend/agents/bdi/bdi_program_manager.py +++ b/src/control_backend/agents/bdi/bdi_program_manager.py @@ -60,24 +60,41 @@ class BDIProgramManager(BaseAgent): await self.send(message) self.logger.debug("Sent new norms and goals to the BDI agent.") + async def _send_clear_llm_history(self): + """ + Clear the LLM Agent's conversation history. + + Sends an empty history to the LLM Agent to reset its state. + """ + message = InternalMessage( + to=settings.agent_settings.llm_name, + sender=self.name, + body="clear_history", + threads="clear history message", + ) + await self.send(message) + self.logger.debug("Sent message to LLM agent to clear history.") + async def _receive_programs(self): """ Continuous loop that receives program updates from the HTTP endpoint. It listens to the ``program`` topic on the internal ZMQ SUB socket. When a program is received, it is validated and forwarded to BDI via :meth:`_send_to_bdi`. + Additionally, the LLM history is cleared via :meth:`_send_clear_llm_history`. """ while True: topic, body = await self.sub_socket.recv_multipart() try: program = Program.model_validate_json(body) + await self._send_to_bdi(program) + await self._send_clear_llm_history() + except ValidationError: self.logger.exception("Received an invalid program.") continue - await self._send_to_bdi(program) - async def setup(self): """ Initialize the agent. diff --git a/src/control_backend/agents/llm/llm_agent.py b/src/control_backend/agents/llm/llm_agent.py index 0263b30..f1c70c9 100644 --- a/src/control_backend/agents/llm/llm_agent.py +++ b/src/control_backend/agents/llm/llm_agent.py @@ -52,6 +52,10 @@ class LLMAgent(BaseAgent): await self._process_bdi_message(prompt_message) except ValidationError: self.logger.debug("Prompt message from BDI core is invalid.") + elif msg.sender == settings.agent_settings.bdi_program_manager_name: + if msg.body == "clear_history": + self.logger.debug("Clearing conversation history.") + self.history.clear() else: self.logger.debug("Message ignored (not from BDI core.") diff --git a/test/unit/agents/bdi/test_bdi_program_manager.py b/test/unit/agents/bdi/test_bdi_program_manager.py index a54360c..573524e 100644 --- a/test/unit/agents/bdi/test_bdi_program_manager.py +++ b/test/unit/agents/bdi/test_bdi_program_manager.py @@ -63,6 +63,7 @@ async def test_receive_programs_valid_and_invalid(): manager = BDIProgramManager(name="program_manager_test") manager.sub_socket = sub manager._send_to_bdi = AsyncMock() + manager._send_clear_llm_history = AsyncMock() try: # Will give StopAsyncIteration when the predefined `sub.recv_multipart` side-effects run out @@ -75,3 +76,24 @@ async def test_receive_programs_valid_and_invalid(): forwarded: Program = manager._send_to_bdi.await_args[0][0] assert forwarded.phases[0].norms[0].norm == "N1" assert forwarded.phases[0].goals[0].description == "G1" + + # Verify history clear was triggered + assert manager._send_clear_llm_history.await_count == 1 + + +@pytest.mark.asyncio +async def test_send_clear_llm_history(mock_settings): + # Ensure the mock returns a string for the agent name (just like in your LLM tests) + mock_settings.agent_settings.llm_agent_name = "llm_agent" + + manager = BDIProgramManager(name="program_manager_test") + manager.send = AsyncMock() + + await manager._send_clear_llm_history() + + assert manager.send.await_count == 1 + msg: InternalMessage = manager.send.await_args[0][0] + + # Verify the content and recipient + assert msg.body == "clear_history" + assert msg.to == "llm_agent" diff --git a/test/unit/agents/llm/test_llm_agent.py b/test/unit/agents/llm/test_llm_agent.py index e2b6460..3341c7d 100644 --- a/test/unit/agents/llm/test_llm_agent.py +++ b/test/unit/agents/llm/test_llm_agent.py @@ -259,3 +259,23 @@ async def test_stream_query_llm_skips_non_data_lines(mock_httpx_client, mock_set # Only the valid 'data:' line should yield content assert tokens == ["Hi"] + + +@pytest.mark.asyncio +async def test_clear_history_command(mock_settings): + """Test that the 'clear_history' message clears the agent's memory.""" + # setup LLM to have some history + mock_settings.agent_settings.bdi_program_manager_name = "bdi_program_manager_agent" + agent = LLMAgent("llm_agent") + agent.history = [ + {"role": "user", "content": "Old conversation context"}, + {"role": "assistant", "content": "Old response"}, + ] + assert len(agent.history) == 2 + msg = InternalMessage( + to="llm_agent", + sender=mock_settings.agent_settings.bdi_program_manager_name, + body="clear_history", + ) + await agent.handle_message(msg) + assert len(agent.history) == 0 diff --git a/test/unit/test_main.py b/test/unit/test_main.py index 2737c76..a423703 100644 --- a/test/unit/test_main.py +++ b/test/unit/test_main.py @@ -53,8 +53,6 @@ async def test_lifespan_agent_start_exception(): Ensures exceptions are logged properly and re-raised. """ with ( - patch("control_backend.main.VADAgent.start", new_callable=AsyncMock), - patch("control_backend.main.VADAgent.reset_stream", new_callable=AsyncMock), patch( "control_backend.main.RICommunicationAgent.start", new_callable=AsyncMock ) as ri_start,