--- title: Agentic video understanding in Gemini | daily.dev description: Instead of feeding an entire video to Gemini as frames (which can exceed 100,000 tokens), an agentic approach lets the model call tools like get_transcript,... canonical: https://daily.dev/posts/agentic-video-understanding-in-gemini-6vw5r9h5n twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Agentic video understanding in Gemini | daily.dev og:description: Instead of feeding an entire video to Gemini as frames (which can exceed 100,000 tokens), an agentic approach lets the model call tools like get_transcript,... og:url: https://daily.dev/posts/agentic-video-understanding-in-gemini-6vw5r9h5n og:image: https://api.daily.dev/og/posts/6vw5r9H5N.png og:image:alt: Agentic video understanding in Gemini og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Agentic video understanding in Gemini **[Google for Developers](https://daily.dev/sources/googledevelopers)** · 3 min read · 0 upvotes · 0 comments ## Summary Instead of feeding an entire video to Gemini as frames (which can exceed 100,000 tokens), an agentic approach lets the model call tools like get_transcript, get_frames, and get_audio to selectively pull only the information needed to answer a query. The model runs a think-act-observe loop, deciding which tools to call and at what granularity (e.g., zooming into a specific timestamp), which cuts token costs while improving accuracy by focusing attention on relevant segments. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why is feeding an entire video as frames to Gemini so expensive? Because passing every frame of a video to Gemini can consume more than 100,000 tokens, even when only a small portion of the video is relevant to the query. This naive approach forces the model to process far more visual context than necessary, driving up cost without improving the quality of the answer. _Anyone optimizing multimodal AI pipelines can track token-cost techniques like this via daily.dev._ ### How does agentic video understanding reduce token costs compared to sending full video frames? It gives the model a reference to the video rather than the whole thing, letting it call tools like get_transcript or get_frames to pull only the segments it needs, such as zooming into frames per second around a specific timestamp. This think-act-observe loop repeats until the model has enough targeted information to answer, avoiding the need to process the entire video. _Developers weighing agentic pipelines against naive context-stuffing can follow patterns like this on daily.dev._ --- Tags: [#google-gemini](https://daily.dev/tags/google-gemini), [#agentic-ai](https://daily.dev/tags/agentic-ai), [#multimodal](https://daily.dev/tags/multimodal) [View this post on daily.dev](https://daily.dev/posts/agentic-video-understanding-in-gemini-6vw5r9h5n) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Agentic video understanding in Gemini","url":"https://daily.dev/posts/agentic-video-understanding-in-gemini-6vw5r9h5n","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/agentic-video-understanding-in-gemini-6vw5r9h5n"},"datePublished":"2026-09-01T17:21:18.259Z","dateModified":"2026-09-02T17:02:31.464Z","description":"Instead of feeding an entire video to Gemini as frames (which can exceed 100,000 tokens), an agentic approach lets the model call tools like get_transcript,...","image":"https://i.ytimg.com/vi/ytjgy30Cono/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/ytjgy30Cono/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Google for Developers","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Google for Developers","logo":"https://media.daily.dev/image/upload/s--jj3nxRoG--/f_auto/v1725035497/logos/googledevelopers","url":"https://daily.dev/sources/googledevelopers"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/agentic-video-understanding-in-gemini-6vw5r9h5n","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"google-gemini,agentic-ai,multimodal","timeRequired":"PT3M","video":{"@type":"VideoObject","name":"Agentic video understanding in Gemini","description":"Instead of feeding an entire video to Gemini as frames (which can exceed 100,000 tokens), an agentic approach lets the model call tools like get_transcript,...","thumbnailUrl":"https://i.ytimg.com/vi/ytjgy30Cono/sddefault.jpg","uploadDate":"2026-09-01T17:21:18.259Z","duration":"PT3M","url":"https://api.daily.dev/r/6vw5r9H5N","embedUrl":"https://www.youtube.com/embed/ytjgy30Cono"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Google for Developers","item":"https://daily.dev/sources/googledevelopers"},{"@type":"ListItem","position":3,"name":"Agentic video understanding in Gemini"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/agentic-video-understanding-in-gemini-6vw5r9h5n#faq","mainEntity":[{"@type":"Question","name":"Why is feeding an entire video as frames to Gemini so expensive?","acceptedAnswer":{"@type":"Answer","text":"Because passing every frame of a video to Gemini can consume more than 100,000 tokens, even when only a small portion of the video is relevant to the query. This naive approach forces the model to process far more visual context than necessary, driving up cost without improving the quality of the answer. Anyone optimizing multimodal AI pipelines can track token-cost techniques like this via daily.dev."}},{"@type":"Question","name":"How does agentic video understanding reduce token costs compared to sending full video frames?","acceptedAnswer":{"@type":"Answer","text":"It gives the model a reference to the video rather than the whole thing, letting it call tools like get_transcript or get_frames to pull only the segments it needs, such as zooming into frames per second around a specific timestamp. This think-act-observe loop repeats until the model has enough targeted information to answer, avoiding the need to process the entire video. Developers weighing agentic pipelines against naive context-stuffing can follow patterns like this on daily.dev."}}]} ```