Asset Search & Intelligence

Runnit’s asset library is a full digital asset manager built for natural-language search. Every file you add is automatically processed so you can find it by meaning, not just by filename, and AI agents can retrieve the right material when they work.

What gets indexed

When a file is uploaded or updated, Runnit processes it for search:

Documents (Markdown, HTML, text) are split into passages and indexed so a search can match the relevant part of a long document.
Images are described automatically. The indexed description covers objects, people, visible text, colours, style, and orientation, so a query like “images featuring a bottle” finds the right pictures. Images also pick up usage tags such as promo-graphic, wide-banner, or social-post.
PDFs have their text extracted and indexed the same way.

Each asset shows an indexing status (pending, processing, ready, or failed) and a short summary in its insights panel. Indexing happens automatically in the background, and a periodic sweep picks up anything that was missed. If processing keeps failing, Runnit waits progressively longer between attempts and eventually retries once a day. Use the re-index action to start a fresh attempt after you fix the cause.

Unmodified starter files supplied by Runnit are not indexed. If an administrator edits one, it becomes ordinary organisation content and is indexed like any other file.

The AI work used to describe and index files appears in AI Usage. For organisations using platform credits, document indexing does not draw from the shared balance. AI Usage still shows the value of that work separately from the credits charged.

Searching the library

Inside a collection, use Search file names to filter the files you can see. Enter any part of a file name; capitalisation does not matter. This works even before a file is indexed. Combine it with the file type filter to narrow the list further. Open a folder to see its matching files. If nothing matches, choose Clear filters to restore all files and file types.

For a search by meaning across your libraries, ask Ru to find what you need. Include the client, campaign, project, folder, tags, file type or date range when useful. For example, ask for promo images from a client’s Summer Launch campaign. Ru searches the files you have permission to access and can show matching passages and images. Structured searches also find files without indexed text, such as videos.

Where the asset browser offers a semantic search field, enter at least three characters to show ranked matches separately from the file list. A filename filter within a collection does not search file contents.

Tags, insights, and versions

Alongside search, each asset carries:

Tags you can add or remove for quick filtering and governance.
Insights: tags, folder, summary, and indexing status in one panel.
Versions: a timeline of every change, with one-click restore to an earlier snapshot.

Indexing status, reindex, and OCR

A file is only searchable by its contents once it has been indexed, and indexing has a quality:

Indexed: the file’s own text, an image description, or an OCR transcript is searchable.
Metadata only: the file can be found by its name and description, but not by what it says. This is usually a scanned document with no text layer.
Scanned, no OCR: a metadata-only document that Runnit can transcribe if you ask it to.
Queued, Indexing, Failed: in progress, or stopped with a reason you can read in the file’s preview.

The preview panel of every file shows its status with a Reindex button and, for scanned documents, a Run OCR button. Above every collection a summary bar shows how many files are searchable, which need attention, and when the collection was last indexed, with Reindex (everything, or only files needing attention) and OCR scanned files actions. Admins see the same for the whole organisation on the Indexing tab of the Organisation Library page, including a background indexing status and a per-collection table.

OCR is opt-in because it uses AI credits: one vision call per scanned document. Turn it on per file, per collection, or for the whole organisation with the “Always OCR scanned documents” setting on the Indexing tab of the Organisation Library page (owners and admins). Runnit reuses a saved transcript when the file content, name and format are unchanged. Edited files need a fresh read. If a saved transcript is unavailable, another transcription uses AI credits.

Reindexing runs in the background and starts immediately. Failed files retry on their own with increasing gaps, so a temporary problem (for example, the organisation being out of AI credits) heals once it is fixed.

How AI agents use it

Runnit’s AI agents search the same library you do. When you ask the in-app assistant something like “show me the approved logos for this client”, it runs the search and answers from the files it found. It fetches the next set of matches if you ask to see more.

The same index is Ru’s knowledge base for plain questions. Every message you send to Ru runs a search of the documents you can access, and Ru reads the matching passages, opens the documents that matter, and answers from them before turning to the web or general knowledge. See Ru answers from your workspace first.

A few safeguards keep this trustworthy:

The assistant only ever links to files that genuinely exist in your library; it cannot invent a file name or a broken link.
When you ask it to create or edit media, it will only claim something was produced if it actually generated it.
If a response starts to promise a lookup or action without doing it (for example “let me check” or “I’m starting now”), Runnit asks the assistant to complete the action in the same response or give the result it already has. Quoted status comments such as “on it” aren’t mistaken for a new promise from the assistant. This includes replies to a confirmation you give: answering “yes” or picking an option runs the work straight away rather than producing another acknowledgement. In the rare case the assistant still ends on an announcement, Runnit appends a correction telling you the work has not started, so you are never left waiting on something that is not running.
Images are shown inline in the chat, and generated videos play directly in an inline player.

See Chat Assistant for page-aware conversations and the confirmed product-feedback workflow.

Finding similar clients

For agencies, the Find similar clients capability compares one client’s brand strategy against your other managed clients to answer questions like “which clients have a similar tone of voice?”, returning a ranked list with the best matching excerpts. This requires the clients’ brand documents to be indexed.

Search only returns files you can read. Expense receipts follow the expense’s current permissions. See receipt access.

Supplier invoice originals need current cost access to every linked purchase record, or organisation cost access when unattached. Use the authenticated original-file download. Public viewer links and collection copies are unavailable; search and indexing keep the same access rules. See Supplier Portal.