diff --git a/docs/concepts/tool_system/twelvelabs_video.md b/docs/concepts/tool_system/twelvelabs_video.md
new file mode 100644
index 00000000..c48b708e
--- /dev/null
+++ b/docs/concepts/tool_system/twelvelabs_video.md
@@ -0,0 +1,95 @@
+# Introduce to TwelveLabs Video tool
+
+# Introduction
+
+[TwelveLabs](https://twelvelabs.io) provides state-of-the-art video understanding models. This tool wraps two of them:
+
+- **Pegasus** (`task: analyze`) — video understanding. Answer a natural language prompt about a video, e.g. summarize it, list the objects shown, or describe what happens.
+- **Marengo** (`task: embed`) — multimodal embeddings. Produce a 512-dim embedding vector from text or an image. Text, image and video share one embedding space, so a text/image vector can be used to retrieve relevant videos.
+
+This lets OmAgent's multimodal agents reason about videos by URL without running a local video model.
+
+# How to use
+
+## Use the TwelveLabs Video tool
+
+YAML file in the configs folder defines available tools. The TwelveLabs Video tool can be enabled by adding the following config.
+
+```yaml
+llm: ${sub|text_res}
+tools:
+    - ...other tools...
+    - name: TwelveLabsVideo # enable the TwelveLabs video tool
+      api_key: ${env|twelvelabs_api_key, null} # set the TwelveLabs API key via environment variable
+```
+
+## Get a TwelveLabs API key
+
+1.  Open [https://twelvelabs.io](https://twelvelabs.io) and sign up (there is a generous free tier).
+
+2.  Create an API key from the dashboard.
+
+3.  Copy it and set the environment variable, e.g. `export twelvelabs_api_key=tlk-xxx` in your terminal, or `os.environ['twelvelabs_api_key'] = "tlk-xxx"` in `run_cli/app/webpage.py`.
+
+## Input Parameters
+
+1.  task
+    1.  type: string
+    2.  enum: ["analyze", "embed"]
+    3.  description: Which capability to use. `analyze` runs Pegasus to answer a prompt about a video. `embed` runs Marengo to produce an embedding vector.
+    4.  required: True
+2.  prompt
+    1.  type: string
+    2.  description: For `analyze`: the question or instruction about the video.
+3.  text
+    1.  type: string
+    2.  description: For `embed`: the text to embed into a multimodal vector.
+4.  image_url
+    1.  type: string
+    2.  description: For `embed`: a public URL to an image to embed.
+5.  video_url
+    1.  type: string
+    2.  description: For `analyze`: a public URL to the video file. TwelveLabs fetches it server-side.
+6.  max_tokens
+    1.  type: integer
+    2.  description: For `analyze`: maximum number of tokens to generate. Default is `2048`.
+
+## Output Data
+
+For `task: analyze`:
+
+1.  text
+    1.  type: string
+    2.  description: The model's answer about the video.
+2.  finish_reason
+    1.  type: string
+    2.  description: Why generation stopped.
+
+For `task: embed`:
+
+1.  embedding
+    1.  type: List[float]
+    2.  description: The 512-dim embedding vector.
+2.  dimension
+    1.  type: integer
+    2.  description: The vector dimension (512).
+
+## Quick Experience
+
+```python
+from omagent_core.tool_system.tools.twelvelabs_video.twelvelabs_video import TwelveLabsVideo
+
+tool = TwelveLabsVideo(api_key="tlk-xxx")
+
+# Marengo embedding
+res = tool.run({"task": "embed", "text": "a cat playing piano"})
+print(res["dimension"], res["embedding"][:3])
+
+# Pegasus video understanding
+res = tool.run({
+    "task": "analyze",
+    "video_url": "https://example.com/your-video.mp4",
+    "prompt": "Describe what happens in this video.",
+})
+print(res["text"])
+```
diff --git a/omagent-core/pyproject.toml b/omagent-core/pyproject.toml
index 8f390aea..56f30420 100644
--- a/omagent-core/pyproject.toml
+++ b/omagent-core/pyproject.toml
@@ -46,6 +46,7 @@ pydub = "^0.25.1"
 qwen_vl_utils = "0.0.8"
 face-recognition = "^1.3.0"
 tavily-python = "^0.5.0"
+twelvelabs = "^1.2.8"
 httpx = {extras = ["http2"], version = "^0.28.1"}
 func-timeout = "^4.3.5"
 gradio = "^5.7.1"
diff --git a/omagent-core/src/omagent_core/tool_system/tools/twelvelabs_video/__init__.py b/omagent-core/src/omagent_core/tool_system/tools/twelvelabs_video/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/omagent-core/src/omagent_core/tool_system/tools/twelvelabs_video/twelvelabs_video.py b/omagent-core/src/omagent_core/tool_system/tools/twelvelabs_video/twelvelabs_video.py
new file mode 100644
index 00000000..d2903936
--- /dev/null
+++ b/omagent-core/src/omagent_core/tool_system/tools/twelvelabs_video/twelvelabs_video.py
@@ -0,0 +1,156 @@
+from typing import Any, Dict
+
+from pydantic import field_validator
+
+from ....utils.logger import logging
+from ....utils.registry import registry
+from ...base import ArgSchema, BaseTool
+
+ARGSCHEMA = {
+    "task": {
+        "type": "string",
+        "enum": ["analyze", "embed"],
+        "description": "Which TwelveLabs capability to use. `analyze` runs the Pegasus model to "
+        "understand a video and answer a prompt (requires `video_url`). `embed` runs the Marengo "
+        "model to produce a multimodal embedding vector for a piece of text or an image "
+        "(requires `text` or `image_url`).",
+        "required": True,
+    },
+    "prompt": {
+        "type": "string",
+        "description": "For `analyze`: the question or instruction about the video, "
+        "e.g. 'Describe what happens in this video' or 'List every product shown'.",
+        "required": False,
+    },
+    "text": {
+        "type": "string",
+        "description": "For `embed`: the text to embed into a multimodal vector "
+        "(shares an embedding space with video, so it can be used for video retrieval).",
+        "required": False,
+    },
+    "image_url": {
+        "type": "string",
+        "description": "For `embed`: a public URL to an image to embed into a multimodal vector.",
+        "required": False,
+    },
+    "video_url": {
+        "type": "string",
+        "description": "For `analyze`: a public URL to the video file. TwelveLabs fetches it server-side.",
+        "required": False,
+    },
+    "max_tokens": {
+        "type": "integer",
+        "description": "For `analyze`: maximum number of tokens to generate. Default is 2048.",
+        "required": False,
+    },
+}
+
+
+@registry.register_tool()
+class TwelveLabsVideo(BaseTool):
+    """Video understanding and multimodal embedding tool backed by the TwelveLabs API.
+
+    Wraps two TwelveLabs models:
+    - Pegasus (``analyze``): video understanding -- answer a natural language prompt about a video.
+    - Marengo (``embed``): produce a 512-dim multimodal embedding vector from text or an image.
+      Text/image/video share one embedding space, so a text or image vector can retrieve videos.
+
+    Get a free API key at https://twelvelabs.io and provide it via ``api_key``
+    (typically ``${env|twelvelabs_api_key}`` in a tool config).
+    """
+
+    class Config:
+        """Configuration for this pydantic object."""
+
+        extra = "allow"
+        arbitrary_types_allowed = True
+
+    args_schema: ArgSchema = ArgSchema(**ARGSCHEMA)
+    description: str = (
+        "Understand videos and generate multimodal embeddings with TwelveLabs. "
+        "Use task='analyze' to answer questions about a video (Pegasus), or "
+        "task='embed' to get an embedding vector for text or an image (Marengo)."
+    )
+    api_key: str
+    analyze_model_name: str = "pegasus1.5"
+    embed_model_name: str = "marengo3.0"
+
+    @field_validator("api_key")
+    @classmethod
+    def api_key_validator(cls, api_key: str) -> str:
+        if not api_key:
+            raise ValueError(
+                "TwelveLabs API key is not provided. Get a free key at https://twelvelabs.io."
+            )
+        return api_key
+
+    def __init__(self, **data: Any) -> None:
+        super().__init__(**data)
+        # Imported here so the dependency is only required when the tool is used.
+        from twelvelabs import TwelveLabs
+
+        self.client = TwelveLabs(api_key=self.api_key)
+
+    def _run(
+        self,
+        task: str,
+        prompt: str = None,
+        text: str = None,
+        image_url: str = None,
+        video_url: str = None,
+        max_tokens: int = 2048,
+    ) -> Dict[str, Any]:
+        if task == "analyze":
+            return self._analyze(
+                prompt=prompt, video_url=video_url, max_tokens=max_tokens
+            )
+        elif task == "embed":
+            return self._embed(text=text, image_url=image_url)
+        else:
+            raise ValueError(
+                "Unknown task {!r}. Must be one of 'analyze' or 'embed'.".format(task)
+            )
+
+    def _analyze(self, prompt: str, video_url: str, max_tokens: int) -> Dict[str, Any]:
+        if not video_url:
+            raise ValueError("`video_url` is required for task='analyze'.")
+        if not prompt:
+            raise ValueError("`prompt` is required for task='analyze'.")
+        from twelvelabs.types.video_context import VideoContext_Url
+
+        try:
+            res = self.client.analyze(
+                model_name=self.analyze_model_name,
+                video=VideoContext_Url(url=video_url),
+                prompt=prompt,
+                max_tokens=max_tokens,
+            )
+            return {"text": res.data, "finish_reason": res.finish_reason}
+        except Exception as e:
+            logging.error(f"TwelveLabs analyze failed: {e}")
+            return {"text": "", "error": str(e)}
+
+    def _embed(self, text: str, image_url: str) -> Dict[str, Any]:
+        if not text and not image_url:
+            raise ValueError(
+                "Either `text` or `image_url` is required for task='embed'."
+            )
+        if text and image_url:
+            raise ValueError(
+                "Provide exactly one of `text` or `image_url` for task='embed', not both."
+            )
+        try:
+            if text:
+                res = self.client.embed.create(
+                    model_name=self.embed_model_name, text=text
+                )
+                vector = res.text_embedding.segments[0].float_
+            else:
+                res = self.client.embed.create(
+                    model_name=self.embed_model_name, image_url=image_url
+                )
+                vector = res.image_embedding.segments[0].float_
+            return {"embedding": vector, "dimension": len(vector)}
+        except Exception as e:
+            logging.error(f"TwelveLabs embed failed: {e}")
+            return {"embedding": [], "error": str(e)}