Update package to latest version and model

agents-core/pyproject.toml

@@ -22,6 +22,7 @@ classifiers = [
 requires-python = ">=3.10"
 dependencies = [
     "getstream[webrtc,telemetry]>=3.1.2,<4",
+    "aiortc>=1.14.0,<1.15.0",
     "python-dotenv>=1.1.1",
     "pillow>=10.4.0",
     "numpy>=1.24.0", # capped at <2.0 via workspace override in root pyproject.toml

plugins/decart/README.md

@@ -22,20 +22,18 @@ This example shows how to use the `RestylingProcessor` to transform a user's vid
 
 ```python
 from vision_agents.core import User, Agent
-from vision_agents.plugins import getstream, openai, decart
+from vision_agents.plugins import getstream, gemini, decart
 
-# Initialize the restyling processor
 processor = decart.RestylingProcessor(
-    initial_prompt="A cute animated movie with vibrant colours",
-    model="mirage_v2"
+    initial_prompt="Studio Ghibli animation style",
+    model="lucy_2_rt",
 )
 
 agent = Agent(
     edge=getstream.Edge(),
     agent_user=User(name="Styled AI"),
-    instructions="You are a helpful assistant.",
-    llm=openai.LLM("gpt-4o-mini"),
-    # Add the processor to the agent's pipeline
+    instructions="Be helpful",
+    llm=gemini.Realtime(),
     processors=[processor],
 )
 ```
@@ -53,6 +51,31 @@ async def change_style(prompt: str) -> str:
     return f"Style changed to: {prompt}"
 ```
 
+### Reference Images ("costumes")
+
+For models like Lucy that accept a reference image, pass it at construction
+time and/or swap it atomically with a prompt via `update_state`:
+
+```python
+processor = decart.RestylingProcessor(
+    model="lucy_2_rt",
+    initial_prompt="A person wearing a superhero costume",
+    initial_image="./costumes/superhero.png",  # your own reference image
+)
+
+# Later — atomically change prompt + reference image
+await processor.update_state(
+    prompt="A person wearing a wizard robe",
+    image="./costumes/wizard.png",
+)
+
+# Image-only update
+await processor.update_state(image=b"<raw image bytes>")
+```
+
+`initial_image` and `update_state(image=...)` accept `bytes`, a local file
+path, an `http(s)` URL, a `data:` URI, or a raw base64 string.
+
 ## Configuration
 
 The plugin requires a Decart API key. You can provide it in two ways:

plugins/decart/example/README.md

@@ -1,16 +1,16 @@
-# Decart Storyteller Example
+# Decart Virtual Try-On Example
 
-This example shows you how to build a real-time storytelling agent using [Vision Agents](https://visionagents.ai/)
-and [Decart](https://decart.ai/). The agent tells a story while transforming your video feed into an animated style that
-matches the narrative.
+This example shows you how to build a real-time virtual try-on ("costume") agent using
+[Vision Agents](https://visionagents.ai/) and [Decart](https://decart.ai/). The agent listens for voice requests like
+"put me in a superhero costume" and uses the Lucy real-time model to restyle the user's video so they appear to be
+wearing it, using a reference image.
 
-In this example, the AI storyteller will:
+In this example, the AI wardrobe assistant will:
 
 - Listen to your voice input
-- Generate a story based on your interactions
-- Use [Decart](https://decart.ai/) to restyle your video feed in real-time (e.g., "A cute animated movie with vibrant
-  colours")
-- Change the video style dynamically as the story progresses
+- Use [Decart](https://decart.ai/) Lucy to restyle your video feed in real-time with both a prompt and a reference image
+- Atomically swap costumes via `processor.update_state(prompt=..., image=...)`
+- Fall back to prompt-only outfit changes for freeform requests
 - Speak with an expressive voice using [ElevenLabs](https://elevenlabs.io/)
 - Run on Stream's low-latency edge network
 
@@ -54,25 +54,36 @@ The agent will:
 1. Create a video call
 2. Open a demo UI in your browser
 3. Join the call
-4. Start telling a story and restyling your video
+4. Listen for costume requests and restyle your video with Lucy
 
 ## Code Walkthrough
 
 ### Setting Up the Agent
 
-The code creates an agent with the Decart processor and other components:
+The code creates an agent with the Decart processor (Lucy real-time) and a pre-defined set of costumes:
 
 ```python
+COSTUMES: dict[str, dict[str, Optional[str]]] = {
+    "jacket": {
+        "prompt": "A person wearing a jacket",
+        "image": "https://images.unsplash.com/photo-1591047139829-d91aecb6caea",
+    },
+    "superhero": {
+        "prompt": "A person wearing a superhero costume",
+        "image": "https://images.unsplash.com/photo-1766062854584-77e3d2467e54",
+    },
+}
+
 processor = decart.RestylingProcessor(
-    initial_prompt="A cute animated movie with vibrant colours",
-    model="mirage_v2"
+    model="lucy_2_rt",
 )
+llm = openai.LLM(model="gpt-5")
 
 agent = Agent(
     edge=getstream.Edge(),
-    agent_user=User(name="Story teller", id="agent"),
-    instructions="You are a story teller...",
-    llm=openai.LLM(model="gpt-4o-mini"),
+    agent_user=User(name="Virtual Wardrobe", id="agent"),
+    instructions="You are a playful virtual wardrobe assistant...",
+    llm=llm,
     tts=elevenlabs.TTS(voice_id="N2lVS1w4EtoT3dr4eOWO"),
     stt=deepgram.STT(),
     processors=[processor],
@@ -81,46 +92,67 @@ agent = Agent(
 
 **Components:**
 
-- `processor`: The Decart RestylingProcessor that transforms the video feed.
-- `llm`: The language model (GPT-4o-mini) that generates the story and controls the processor.
-- `tts`: ElevenLabs TTS for expressive voice output.
-- `stt`: Deepgram STT for transcribing user speech.
-- `processors`: The list of video processors (just Decart in this case).
+- `processor`: The Decart `RestylingProcessor` running the `lucy_2_rt` real-time model, which accepts a reference image.
+- `llm`: GPT-5 — picks the right costume and narrates the change.
+- `tts` / `stt`: ElevenLabs + Deepgram for a voice-driven loop.
 
-### Dynamic Style Changing
+### Swapping Costumes Atomically
 
-The agent can change the video style dynamically using a registered function:
+`update_state` mirrors the JS SDK's `realtimeClient.set({ prompt, enhance, image })` — prompt and reference image are
+applied in a single atomic update so the output video never shows a half-updated state:
 
 ```python
 @llm.register_function(
-    description="This function changes the prompt of the Decart processor which in turn changes the style of the video and user's background"
+    description="Put the user in one of the pre-defined costumes."
 )
-async def change_prompt(prompt: str) -> str:
-    await processor.update_prompt(prompt)
-    return f"Prompt changed to {prompt}"
+async def change_costume(name: str) -> str:
+    costume = COSTUMES.get(name.lower())
+    if costume is None:
+        return f"Unknown costume '{name}'. Available: {', '.join(COSTUMES)}."
+    await processor.update_state(prompt=costume["prompt"], image=costume["image"])
+    return f"Costume changed to {name}."
 ```
 
-This allows the LLM to call `change_prompt("A dark and stormy night")` to instantly change the visual style of the video
-to match the story's mood.
+For freeform requests (anything not in `COSTUMES`), the agent calls `change_outfit` which uses
+`update_state(prompt=..., image=...)` if the user supplies a URL, or `update_prompt(...)` for prompt-only changes:
+
+```python
+@llm.register_function(
+    description=(
+        "Change the user's outfit to a freeform description. Use this when "
+        "the user asks for a costume not in the pre-defined list. If you "
+        "have a reference image URL (http/https) pass it as image_url, "
+        "otherwise pass an empty string."
+    )
+)
+async def change_outfit(description: str, image_url: str) -> str:
+    if image_url:
+        await processor.update_state(prompt=description, image=image_url)
+    else:
+        await processor.update_prompt(description)
+    return f"Outfit changed: {description}"
+```
 
 ## Customization
 
-### Change the Initial Style
+### Add or Change Costumes
+
+Edit the `COSTUMES` dict. Each entry needs a `prompt` and an optional `image` — bytes, a file path, an http(s) URL, a
+data URI, or a raw base64 string are all accepted.
+
+### Start With a Costume Already On
 
-Modify the `initial_prompt` in the `RestylingProcessor` to start with a different look:
+Pass `initial_image` to the processor so the very first frame is already restyled. Point it at your own hosted image (or
+a local file path / bytes / data URI):
 
 ```python
 processor = decart.RestylingProcessor(
-    initial_prompt="A cyberpunk city with neon lights",
-    model="mirage_v2"
+    model="lucy_2_rt",
+    initial_prompt="A person wearing a superhero costume",
+    initial_image="./costumes/superhero.png",  # or bytes, an http(s) URL, a data URI, or raw base64
 )
 ```
 
-### Modify the Storytelling Persona
-
-Edit the `instructions` passed to the `Agent` to change the storyteller's personality, tone, or the type of stories they
-tell.
-
 ### Change the Voice
 
 Update the `voice_id` in `elevenlabs.TTS` to use a different ElevenLabs voice.

plugins/decart/example/decart_example.py

@@ -1,4 +1,5 @@
 import logging
+from typing import Optional
 
 from dotenv import load_dotenv
 from vision_agents.core import Agent, Runner, User
@@ -9,46 +10,92 @@
 
 load_dotenv()
 
+# Pre-defined outfits for the virtual try-on / "costume" demo. The user asks the
+# agent to put them in one and the LLM calls change_costume(...) which
+# atomically updates the prompt + reference image on the Decart Lucy model.
+#
+# To enable the reference-image ("virtual try-on") feature, set `image` to your
+# own hosted reference image. Any of the following are accepted:
+#   - bytes
+#   - a local file path (e.g. "./costumes/superhero.png")
+#   - an http(s) URL
+#   - a data: URI
+#   - a raw base64 string
+# When `image` is None the prompt alone drives the restyling.
+COSTUMES: dict[str, dict[str, Optional[str]]] = {
+    "jacket": {
+        "prompt": "A person wearing a jacket",
+        "image": "https://images.unsplash.com/photo-1591047139829-d91aecb6caea",
+    },
+    "superhero": {
+        "prompt": "A person wearing a superhero costume",
+        "image": "https://images.unsplash.com/photo-1766062854584-77e3d2467e54",
+    },
+}
+
 
 async def create_agent(**kwargs) -> Agent:
     processor = decart.RestylingProcessor(
-        initial_prompt="A cute animated movie with vibrant colours", model="mirage_v2"
+        model="lucy_2_rt",
     )
-    llm = openai.LLM(model="gpt-4o-mini")
+    llm = openai.LLM(model="gpt-5")
 
     agent = Agent(
         edge=getstream.Edge(),
-        agent_user=User(name="Story teller", id="agent"),
-        instructions="You are a story teller. You will tell a short story to the user. You will use the Decart processor to change the style of the video and user's background. You can embed audio tags in your responses for added effect Emotional tone: [EXCITED], [NERVOUS], [FRUSTRATED], [TIRED] Reactions: [GASP], [SIGH], [LAUGHS], [GULPS] Volume & energy: [WHISPERING], [SHOUTING], [QUIETLY], [LOUDLY] Pacing & rhythm: [PAUSES], [STAMMERS], [RUSHED]",
+        agent_user=User(name="Virtual Wardrobe", id="agent"),
+        instructions=(
+            "You are a playful virtual wardrobe assistant. The user is on a "
+            "live video call and you can change what they appear to be wearing "
+            "in real time by calling change_costume(name) with one of the "
+            f"available costumes: {', '.join(COSTUMES)}. If the user asks for a "
+            "costume that isn't in that list, call change_outfit(description, "
+            "image_url) instead. Describe each transformation out loud in a "
+            "single short sentence. You can embed audio tags for effect, e.g. "
+            "[sigh], [excited], [pause], [rushed], or [tired]"
+        ),
         llm=llm,
         tts=elevenlabs.TTS(voice_id="N2lVS1w4EtoT3dr4eOWO"),
         stt=deepgram.STT(),
         processors=[processor],
     )
 
     @llm.register_function(
-        description="This function changes the prompt of the Decart processor which in turn changes the style of the video and user's background"
+        description=("Put the user in one of the pre-defined costumes.")
     )
-    async def change_prompt(prompt: str) -> str:
-        await processor.update_prompt(prompt)
-        return f"Prompt changed to {prompt}"
+    async def change_costume(name: str) -> str:
+        costume = COSTUMES.get(name.lower())
+        if costume is None:
+            return f"Unknown costume '{name}'. Available: {', '.join(COSTUMES)}."
+        await processor.update_state(prompt=costume["prompt"], image=costume["image"])
+        return f"Costume changed to {name}."
+
+    @llm.register_function(
+        description=(
+            "Change the user's outfit to a freeform description. Use this when "
+            "the user asks for a costume not in the pre-defined list. If you "
+            "have a reference image URL (http/https) pass it as image_url, "
+            "otherwise pass an empty string."
+        )
+    )
+    async def change_outfit(description: str, image_url: str) -> str:
+        if image_url:
+            await processor.update_state(prompt=description, image=image_url)
+        else:
+            await processor.update_prompt(description)
+        return f"Outfit changed: {description}"
 
     return agent
 
 
 async def join_call(agent: Agent, call_type: str, call_id: str, **kwargs) -> None:
     """Join the call and start the agent."""
-    # Create a call
     call = await agent.create_call(call_type, call_id)
-
     logger.info("🤖 Starting Agent...")
 
-    # Have the agent join the call/room
     async with agent.join(call):
-        logger.info("Joining call")
-        logger.info("LLM ready")
+        await agent.simple_response(text="Hello! Tell me what you can do.")
 
-        await agent.finish()  # Run till the call ends
+        await agent.finish()
 
 
 if __name__ == "__main__":

plugins/decart/pyproject.toml

@@ -12,7 +12,7 @@ requires-python = ">=3.10"
 license = "MIT"
 dependencies = [
     "vision-agents",
-    "decart==0.0.8",
+    "decart==0.0.29",
 ]
 
 [project.urls]