Searchunk

Searchunk is an AI-powered search layer for image and video archives. It reads your archive with AI wherever it lives — in cloud storage such as Amazon S3, Google Cloud Storage or Azure Blob Storage, or in folders on your own computer — and lets you find any frame by describing it in your own words. Instead of relying on file names, folder structures or hand-typed tags, you type a phrase such as "dog running on the beach" and Searchunk returns the best matching images and video frames in seconds. The product is aimed at companies and teams that keep large photo and video collections and need to retrieve a specific shot quickly. Searches work in both English and Turkish, and the original files never leave the storage the customer already uses. Most media archives are not organized around the way people actually remember images. A photo might be called IMG_4471.HEIC, DSC08812.JPG, P1030127.jpg or export-final.webp, and the moment someone needs it — a sign outside a veterinary clinic, a dog on a beach, a particular scene in a video — the file name offers no help. Folder structures capture only the broadest categories, and hand-typed tags require someone to sit down and label thousands of files in advance, which rarely happens consistently and almost never keeps pace as new material arrives. Video makes the problem sharper, because a single clip can hold dozens of distinct scenes and the relevant one may last only a few seconds. Searchunk addresses this by indexing what is actually visible in each image and each scene of each video, so retrieval depends on content rather than on how someone once named a file. A large part of an archive's meaning is written rather than photographed. Searchunk reads the text in images, so signs, labels, posters and screens become searchable. The website illustrates this with a frame showing a green veterinary clinic sign reading "SAHİL VETERİNER": typing that phrase brings the frame up, even though the file itself gives no hint of the sign. Alongside the text, the AI records what it sees in a frame — the example lists a vet clinic, a shop front, a green sign and a paw print. Searchunk also understands Turkish, including endings: a search for "kuşlar" also finds files described as "muhabbet kuşu", and typing Turkish words without Turkish letters still works. Videos are split into scenes and each scene is searched on its own. Results show timestamps such as 00:42 next to a matching clip, and clicking a result starts the video playing right at that moment rather than at the beginning. Combined with the ranking behaviour — the best frames come first, with visible match scores such as 94% or 41% — this turns a long recording into something closer to a set of individually findable shots. Because each scene is treated separately, a specific few seconds inside a longer MP4, MOV, WEBM, MKV, AVI or M4V file can be located directly by describing it, and the results page can be filtered between all results, images and videos. Searchunk connects to wherever the archive already lives instead of requiring it to be uploaded somewhere new. Supported sources include Amazon S3, Google Cloud Storage, Azure Blob Storage, OVHcloud, Wasabi, Backblaze B2, Cloudflare R2, DigitalOcean Spaces, Scaleway, Hetzner, Akamai (Linode), Exoscale SOS, IONOS, Contabo, Oracle Cloud, IBM Cloud, IDrive e2 and any S3-compatible storage, plus folders on your computer via a small agent for Windows, macOS or Linux. Setup takes a few minutes. After that, sources are checked regularly: new, changed and deleted files are handled automatically, so new files join the indexing queue on their own and the archive does not need to be re-submitted each time it grows. The workflow the website describes has three steps. First you connect your archive by selecting a storage service or installing the agent for a local folder; your files stay where they are. Second, Searchunk's AI looks at every image and every scene of your videos, capturing objects, the setting, the colours and the text on signs. Third, you search by describing what you have in mind, in English or Turkish, and the best frames come first. Every search runs in both languages. Behind this sits a semantic search model: content is indexed up front and queries are matched against what was seen, not against metadata. Because the product is an index rather than a hosting service, it does not take over the archive. Originals are not copied; only search data and small previews are kept, and the agent on a computer sends only downscaled frames. Every AI model runs on Searchunk's own servers, so images are never sent to outside AI services. Storage credentials are stored encrypted, and each company's data is kept apart and never mixes with another company's searches. A privacy notice explains which data is processed and why. Practically, this means a team gets fast, description-based retrieval without moving large amounts of media, without changing its storage provider, and without exposing its archive to third-party AI vendors. Searchunk is built for teams as well as individuals. Colleagues at a company's domain can be invited, and everyone searches the same archive. For integration into other tools, an API key unlocks the same search from an application or workflow; the documentation shows a call to the searchunk.com API endpoint with a bearer token and a JSON body containing a query such as "dog running on the beach". Accounts are opened with a company e-mail address — addresses from free services such as Gmail, Hotmail or Outlook are not accepted. Typical use is retrieval inside an existing collection: a team member remembers a sign reading "Sahil Veteriner" and finds the frame without knowing the file name; an editor needs the moment at 00:42 of a long video and jumps straight to it; a photographer searches a personal folder on their computer for a dog running on a beach. Because the same search is available through the API, the capability can be embedded in internal tools or other workflows, and because new, changed and deleted files are handled automatically, the archive stays searchable as it evolves without manual re-tagging. Searchunk indexes images in JPG, PNG, WEBP, AVIF, HEIC, GIF, BMP and TIFF, and videos in MP4, MOV, WEBM, MKV, AVI and M4V; documents such as PDFs are not indexed. Every new company account indexes its first 250 images and videos free, with no setup fee and no credit card required, and every feature is included, including inviting teammates. Files beyond the limit are not indexed while the quota stands, but searches keep working; a higher quota can be requested from the dashboard, and indexing picks up where it stopped once it is raised. The service is currently in public beta, so occasional outages and changes are expected. Searchunk's value proposition is narrow and practical: that frame in your archive is one sentence away. By indexing images and video scenes with AI, keeping originals in the customer's own storage, running every model on its own infrastructure and accepting natural language in English and Turkish, it removes the dependency on file names, folder hierarchies and manual tags. Teams connect a source, describe what they need, and get the best frames first — without moving their archive or changing where it lives.