feat: add director notes support (initial config + runtime cues)

README.md

@@ -152,7 +152,7 @@ async with client.connect(session_options=session_options) as session:
 The main client class for connecting to Anam AI.
 
 ```python
-from anam import AnamClient, PersonaConfig, ClientOptions
+from anam import AnamClient, ClientOptions, DirectorNotes, PersonaConfig
 
 # Simple initialization for pre-defined personas - all other parameters are ignored except enable_audio_passthrough
 client = AnamClient(
@@ -172,6 +172,10 @@ client = AnamClient(
         system_prompt="You are a helpful assistant...",
         avatar_model="cara-4",
         language_code="en",
+        director_notes=DirectorNotes(
+            preset_style="warm",
+            expressivity=0.7,
+        ),
         enable_audio_passthrough=False,
     ),
 )
@@ -300,6 +304,9 @@ async with client.connect() as session:
 
     # Interrupt the avatar if speaking
     await session.interrupt()
+
+    # Runtime director-note cue (turn-scoped)
+    await session.send_director_note_cue("curious", at_seconds=0.5)
 
     # Get message history
     history = client.get_message_history()
@@ -310,6 +317,43 @@ async with client.connect() as session:
     await session.wait_until_closed()
 ```
 
+### Director Notes
+
+Director notes control how the avatar performs a conversation (how something is said, not just what is said). Set session defaults on `PersonaConfig(director_notes=...)`, then send runtime `director_note_cue` messages with `session.send_director_note_cue(...)`.
+For supported styles/cues and timing semantics (`at_seconds` vs `in_seconds`), see the official docs: [Director Notes](https://anam.ai/docs/personas/director-notes).
+
+#### Initial Director Notes
+
+Set director notes directly on `PersonaConfig` to apply them from session start:
+
+```python
+from anam import DirectorNotes, PersonaConfig
+
+persona = PersonaConfig(
+    avatar_id="your-avatar-id",
+    director_notes=DirectorNotes(
+        preset_style="warm",
+        expressivity=0.7,
+    ),
+)
+```
+The initial director notes are the avatar's default presence for the session.
+
+##### Runtime Director Note Cues
+
+Use `session.send_director_note_cue(...)` to adjust delivery style during the active turn.
+
+```python
+import asyncio
+
+async with client.connect() as session:
+    cues = ["warm", "playful", "concerned", "laughter"]
+
+    for cue in cues:
+        await session.send_director_note_cue(cue, at_seconds=3.0)
+        await asyncio.sleep(8)
+```
+
 ## Examples
 
 ### Save Video and Audio
@@ -481,7 +525,7 @@ Ephemeral personas give you full control over components at startup. Configure a
 They are ideal for production environments where you need to control the components at startup.
 
 ```python
-from anam import PersonaConfig
+from anam import DirectorNotes, PersonaConfig
 
 # Ephemeral: specify avatar_id, voice_id, and optionally llm_id, avatar_model
 persona = PersonaConfig(
@@ -490,6 +534,10 @@ persona = PersonaConfig(
     llm_id="your-llm-id",             # From https://lab.anam.ai/llms (optional)
     avatar_model="cara-4",            # Video frame model (optional)
     system_prompt="You are...",       # See https://docs.anam.ai/concepts/prompting-guide
+    director_notes=DirectorNotes(     # Optional: Initial director notes 
+        preset_style="warm",
+        expressivity=0.7,
+    ),
     enable_audio_passthrough=False,
 )
 ```

src/anam/__init__.py

@@ -59,6 +59,7 @@ async def consume_audio():
     AnamEvent,
     ClientOptions,
     ConnectionClosedCode,
+    DirectorNotes,
     EgressDailyOptions,
     EgressOptions,
     Message,
@@ -82,6 +83,7 @@ async def consume_audio():
     "AudioFrame",
     "ClientOptions",
     "ConnectionClosedCode",
+    "DirectorNotes",
     "EgressDailyOptions",
     "EgressOptions",
     "Message",

src/anam/client.py

@@ -3,7 +3,9 @@
 from __future__ import annotations
 
 import asyncio
+import json
 import logging
+import math
 import uuid
 from collections.abc import AsyncIterator
 from typing import Any, Awaitable, Callable, TypeVar
@@ -531,9 +533,6 @@ async def send_message(self, content: str) -> None:
         Raises:
             SessionError: If not connected or if LLM is not available.
         """
-        if not self._client._streaming_client:
-            raise SessionError("Not connected")
-
         # Validate that LLM is available for processing messages
         persona_config = self._get_persona_config()
 
@@ -545,13 +544,7 @@ async def send_message(self, content: str) -> None:
                 "Persona ID and LLM ID are not set, messages will not be processed by the backend."
             )
 
-        # Wait for data channel to be ready
-        streaming = self._client._streaming_client
-        if not getattr(streaming, "_data_channel_open", False):
-            logger.debug("Waiting for data channel to open...")
-            if not await streaming.wait_for_data_channel(timeout=10.0):
-                raise SessionError("Data channel did not open in time")
-
+        streaming = await self._ensure_data_channel_open()
         streaming.send_user_message(content)
 
     async def interrupt(self) -> None:
@@ -560,17 +553,57 @@ async def interrupt(self) -> None:
         Raises:
             SessionError: If not connected.
         """
+        streaming = await self._ensure_data_channel_open()
+        streaming.send_interrupt()
+
+    async def send_director_note_cue(
+        self,
+        tag: str,
+        *,
+        at_seconds: float | None = None,
+        in_seconds: float | None = None,
+    ) -> None:
+        """Send a director-note cue over the active session data channel.
+
+        Args:
+            tag: Director-note cue tag (for example ``"curious"``).
+            at_seconds: Optional turn-relative cue time in seconds.
+            in_seconds: Optional delay relative to receipt time in seconds.
+
+        Raises:
+            SessionError: If not connected, data channel is unavailable, or send fails.
+            ValueError: If a provided timing value is not finite.
+        """
+        for field_name, value in (("at_seconds", at_seconds), ("in_seconds", in_seconds)):
+            if value is not None and not math.isfinite(value):
+                raise ValueError(f"{field_name} must be a finite number")
+
+        payload: dict[str, Any] = {
+            "message_type": "director_note_cue",
+            "cue": {"tag": tag},
+        }
+
+        if at_seconds is not None:
+            payload["at_seconds"] = at_seconds
+
+        if in_seconds is not None:
+            payload["in_seconds"] = in_seconds
+
+        streaming = await self._ensure_data_channel_open()
+        if not streaming.send_data_message(json.dumps(payload)):
+            raise SessionError("Failed to send director note cue over data channel")
+
+    async def _ensure_data_channel_open(self):
+        """Return the streaming client once the data channel is ready."""
         if not self._client._streaming_client:
             raise SessionError("Not connected")
 
-        # Wait for data channel to be ready (same as send_message)
         streaming = self._client._streaming_client
         if not getattr(streaming, "_data_channel_open", False):
             logger.debug("Waiting for data channel to open...")
             if not await streaming.wait_for_data_channel(timeout=10.0):
                 raise SessionError("Data channel did not open in time")
-
-        streaming.send_interrupt()
+        return streaming
 
     def create_talk_stream(self, correlation_id: str | None = None) -> TalkMessageStream:
         """Create a talk message stream for sending text chunks to TTS.

src/anam/types.py

@@ -47,6 +47,34 @@ class MessageRole(str, Enum):
     SYSTEM = "system"
 
 
+@dataclass
+class DirectorNotes:
+    """Director-notes settings applied at session start via ``PersonaConfig``.
+
+    These fields are optional and serialized only when set.
+    """
+
+    custom_style_prompt: str | None = None
+    preset_style: str | None = None
+    idling_prompt: str | None = None
+    overlay_model_name_on_output: bool | None = None
+    expressivity: float | None = None
+
+    def to_dict(self) -> dict[str, Any]:
+        result: dict[str, Any] = {}
+        if self.custom_style_prompt is not None:
+            result["customStylePrompt"] = self.custom_style_prompt
+        if self.preset_style is not None:
+            result["presetStyle"] = self.preset_style
+        if self.idling_prompt is not None:
+            result["idlingPrompt"] = self.idling_prompt
+        if self.overlay_model_name_on_output is not None:
+            result["overlayModelNameOnOutput"] = self.overlay_model_name_on_output
+        if self.expressivity is not None:
+            result["expressivity"] = self.expressivity
+        return result
+
+
 @dataclass
 class PersonaConfig:
     """Configuration for an Anam persona.
@@ -62,6 +90,7 @@ class PersonaConfig:
         llm_id: LLM model to use (optional). Set to 'CUSTOMER_CLIENT_V1' to disable
             Anam's default brain for custom LLM integration.
         max_session_length_seconds: Maximum session duration (optional).
+        director_notes: Optional director-notes defaults applied at session start.
         enable_audio_passthrough: If True, bypasses Anam's orchestration layer
             and allows to ingest TTS audio directly through the socket.
     """
@@ -75,6 +104,7 @@ class PersonaConfig:
     language_code: str | None = None
     llm_id: str | None = None
     max_session_length_seconds: int | None = None
+    director_notes: DirectorNotes | None = None
     enable_audio_passthrough: bool | None = False
 
     def to_dict(self) -> dict[str, Any]:
@@ -98,6 +128,10 @@ def to_dict(self) -> dict[str, Any]:
             result["llmId"] = self.llm_id
         if self.max_session_length_seconds is not None:
             result["maxSessionLengthSeconds"] = self.max_session_length_seconds
+        if self.director_notes is not None:
+            director_notes = self.director_notes.to_dict()
+            if director_notes:
+                result["directorNotes"] = director_notes
         if self.enable_audio_passthrough is not None:
             result["enableAudioPassthrough"] = self.enable_audio_passthrough
         return result