> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tessact.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# AI Search
> Find the exact moment across your library by describing scenes, shots, spoken moments, people, settings, and visual cues in plain language.
## Overview
AI Search finds moments by meaning, not by filename. Describe what you remember (a scene, a shot, a spoken line, a person, a mood) and Tessact returns the matching moments from across your library, each with a thumbnail, a time range, and a one-click path into the asset or the Editor.
* **One search surface:** a single query runs across video, image, audio, and document assets in the workspace.
* **Moments, not files:** video results are shots, scenes, and spoken spans with usable in and out points, ready to preview or send to the Editor.
* **Multilingual dialogue:** spoken-word search works across languages, including Hindi and Hinglish.
* **Understands intent:** the query is interpreted before it is matched; AI Search even adjusts loose wording and tells you when it did.
***
## Open AI Search
Select **AI Search** at the very top of the left sidebar, above your Productivity links. You can also start from the **Library**: expand the search field in the action bar, type a query, and press Enter; the search runs scoped to the folder you were in.
The landing page cycles example prompts in the search bar, suggests queries beneath it, and keeps your **Recent searches** one click away.
***
## Run a search
Type what you would say to a colleague: the subject, the setting, an action, or a line of dialogue. You do not need the filename. You can also record a spoken query with the microphone button instead of typing.
Use the controls under the search bar:
| Control | What it does |
| ----------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Search inside** | Opens a Library picker so the search runs only inside a chosen folder or file |
| **Search All** | Chooses which result types come back: **Shot**, **Scene**, **Spoken**, **Image**, **Documents** |
| **Match by** | Steers the matching: **Best match** (default), **Visuals** for what appears on screen, **Transcript** for dialogue, narration, and transcripts, or **Context** for what the content is about |
Results appear under an **AI Results** header with a running total, such as *showing 20 out of 47 results*, and the result-type chips let you show or hide **Shot**, **Scene**, **Spoken**, **Image**, and **Documents** results without rerunning the search. If AI Search rewrote loose wording into a stronger query, a **Query adjusted to** note shows exactly what was searched, so nothing happens silently.
Phrase queries as *subject + setting + action*, e.g. *friends catching up over drinks at a restaurant*. If a title, show, or person name is part of the query, Tessact uses it to scope the search automatically.
***
## What can you search for?
If you can describe it, you can find it. During analysis Tessact indexes what is visible, what is said, how it is filmed, and what it all means, so every one of these query styles works. Steal these and adapt them to your library:
* `Abhilash apologises to Guri for their falling out`
* `two friends arguing until one walks out`
* `a sincere conversation between old friends at night`
* `crowd celebrating and hugging`
Names work when people are identified in your content; emotions and interactions work everywhere.
* `roadside tea stall at night`
* `dimly lit restaurant, indoor`
* `misty hills at sunrise`
* `busy street in the rain`
Settings, indoor or outdoor, weather, and time of day are all part of the index.
* `friends raise a toast`
* `someone reads a message on their phone and reacts`
* `classroom lecture that frames preparation as a marathon`
* `flashback to an older memory at the bar`
Describe what happens, not what the file is called.
* `extreme close-up of a worried face`
* `wide establishing shot of the city`
* `over-the-shoulder conversation`
* `low-angle shot walking towards camera`
Shot size, angle, and framing are detected per shot, so you can search the way an editor talks.
* `the line about turning pressure into a lifestyle`
* `dialogue where they debate the optional subject`
* `the joke about SK's haircut`
* `"tension mat le" moments in Hinglish`
You need the idea, not the exact wording, and Hindi or Hinglish dialogue is searchable too.
* `scenes where a brand logo appears on screen`
* `frames showing a WhatsApp notification`
* `the passage about pricing in the product PDF`
* `poster image with a red background`
On-screen text, logos, document passages, and image styles are all fair game.
These dimensions combine naturally: `medium shot of three friends debating the exam at a tea stall at night` mixes framing, people, action, setting, and time in one query.
***
## Work with the results
Every result card carries the evidence you need to judge it at a glance: a thumbnail of the exact moment (document results preview their matching pages), a type badge such as **Scene**, the clip duration, a descriptive title, and a short summary of what happens. Hovering a card reveals its quick actions: **Add selection** to start collecting clips, and an expand control that jumps straight into the full preview.
### Preview a result
Select a card to open it full screen. The matched segment plays immediately, the header names the source (show, episode, and file), and the moment's full profile sits beside the player.
The header actions cover everything you would do next with a result:
| Control | What it does |
| -------------------------- | -------------------------------------------------------------------------------------------------- |
| **Copy link** (chain icon) | Copies a shareable link to this exact result, so a teammate lands on the same moment |
| **Download** (arrow icon) | Downloads the matched media to your machine |
| **View asset** | Opens the source asset in the Library at the matched moment, with comments and metadata around it |
| **Open in editor** | Sends the asset to the Tessact Editor with the matched range carried over, ready to trim and place |
### The moment's profile
The panel beside the player explains why this moment matched and what is in it, section by section:
* **Output type:** what kind of result you are looking at, such as *Scene*.
* **Summary:** an AI-written account of what actually happens in the clip, detailed enough to judge without scrubbing.
* **Keywords:** thematic tags like *#friendship* or *#reconciliation* for quick topical context.
* **Settings:** where and when the moment takes place, such as *lounge*, *night*, *indoor*.
* **Emotions:** the mood detected in the moment, such as *sincere*.
* **Objects:** notable items visible in frame, such as *glass*, *table*, *suit jacket*.
### Compare and give feedback
The strip along the bottom keeps the rest of your results one click away: use the thumbnail rail or the arrow buttons to flip between matches without leaving the preview. Next to it, **How was this search result?** takes a quick thumbs up or down; feedback tunes ranking for your workspace over time.
***
## From search to edit, in one move
This is the flow editors love: describe the moments you need, tick them off the results, and land in the Editor with every clip already on hand. No hunting through bins, no exporting and re-importing.
Hover any result and select **Add selection** (the **+** control). A checkmark marks it as picked, and you can keep collecting across the whole result set, mixing scenes, shots, and even document moments.
The floating bar at the bottom counts what you have picked, for example **2 files selected**, sourced *From Library*. Clear everything with the **X** if you want to start over.
Select the draft control on the bar to open **Edit selected clips**. The dialog notes that *Editing these clips creates a Video Draft*: choose where it lives with **Location: In Library**, then select **Open in editor**.
Tessact creates the Video Draft and opens it in the Editor with your searched clips ready on the timeline.
From a described moment to an editable timeline without ever browsing folders.
***
## Refine your query
* **Refine one axis at a time:** add a time of day, switch **Match by**, or drop one adjective; small changes keep results interpretable.
* **Broaden when empty:** remove the least essential qualifier, reset **Search All**, and widen the **Search inside** scope before rewriting the whole query.
* **Trust the adjustment note:** if **Query adjusted to** appears, the rewritten phrasing is a good starting point for your next refinement.
Newly uploaded assets become searchable once their AI analysis completes; a livestream becomes searchable chunk by chunk while it is still live. If a search reports no indexed assets in the current scope, the content there has not been analysed yet.
***
## Related guides
Organise the media that AI Search draws from.
Live sessions are indexed while they run, so you can search a stream before it ends.
Pull searched moments straight onto the timeline while editing.
***
You can now find any moment in your library by describing it, steer the search with scopes and filters, and jump from a result straight into review or editing.