Find the right moment in a video by describing it

TwelveLabsTools now supports Marengo embeddings. Marengo embeds text into the same latent space TwelveLabs uses for video, audio, and image, so a written query and a video clip come out as vectors you can compare directly. You describe what you're after, and the embedding lands right next to the moments that match it.‍

from agno.agent import Agent
from agno.tools.twelvelabs import TwelveLabsTools

agent = Agent(
    tools=[TwelveLabsTools(enable_embed_text=True, enable_analyze_video=False)],
    markdown=True,
)

agent.print_response("Embed 'a cat playing piano' so I can search my video index for it")

‍

This is what you need to search a video library by meaning rather than metadata. Embed the query with Marengo, compare it against your indexed video embeddings, and get back the clips that actually show what you asked for instead of the ones that happen to name it in a title or transcript.

The toolkit's other half covers the single-video case. analyze_video uses the Pegasus model to answer natural-language questions about one clip, so an agent can both dig into a specific video and pull the right one out of a corpus of thousands.

View the TwelveLabs toolkit docs to learn more.