Find files by what is in them
Search your photos and documents by what they are about and by the text inside them, not only by file name. Cloudflare Workers AI does the processing, and the index stays in your workspace's region.
Paid plans · off by default · turn it on per workspace
What you can find
PDF and DOCX documents, and JPEG, PNG, WebP and HEIC photos. File name search works as it always has, alongside.
-
By meaning
Describe what you want in plain words. "beach" finds beach photos whatever the files are called, and "the contract with the cancellation clause" finds the PDF that has one.
-
Exact text
A receipt total, a word on a sign, a reference number in a scanned letter. Text in documents and in photos is searchable as written, once Cloudflare Workers AI has processed them. It has to be legible: small, blurred or handwritten text may not come through.
-
Place and date
Each photo's capture date and location from its EXIF data, the same data the Photo Map uses, is stored with its index entry. The search API takes a date range and a map area, and the Claude and ChatGPT connector can ask for a date range. The search boxes in the apps do not offer these filters yet.
How it works
Three steps, and only the first one is yours.
- 1
The owner turns it on
In Settings → AI search, the workspace owner, or someone whose role has the Change AI search permission, switches on "Search inside files and photos". "Also index existing files" covers what is already there, and "Exclude folders" keeps chosen folders out.
- 2
Cloudflare Workers AI processes the files
New and existing files are sent automatically to Cloudflare Workers AI: bge-m3 turns text into search vectors, and Llama 3.2 Vision writes a short description, tags and the visible text of each photo. The result is written to a search index in the workspace's own region.
- 3
Search from anywhere you already search
The web app's search box shows an "In file contents" group under the name matches. The desktop app's Search page, from version 3.1.6, shows the same results, and the Claude and ChatGPT connector has a search_content tool.
FilesSee all
In file contents
Limits by plan
Included in every paid plan. Photo, document and text limits apply to each workspace; processing units belong to the account that pays.
| Plan | Photos | Documents | Document text | Units for existing files | Units each month |
|---|---|---|---|---|---|
| Starter | 10,000 | 2,000 | ≈ 6 million words | 45,000 | 6,000 |
| Plus | 50,000 | 10,000 | ≈ 15 million words | 150,000 | 20,000 |
| Pro | 150,000 | 20,000 | ≈ 30 million words | 375,000 | 50,000 |
| Business | 250,000 | 25,000 | ≈ 45 million words | 600,000 | 80,000 |
| 7-day trial | 200 | 100 | ≈ 600,000 words | 2,200 | None |
- A processing unit is one photo, or about 4,000 characters of document text (roughly 600 English words). Free has no Smart Search.
- Document text is approximate: the index keeps documents in sections of about 300 English words each.
- Per file: documents up to 25 MB; each is indexed up to 300 pages, and up to about 800,000 characters of text.
What happens at a limit
- Indexing pauses New files wait instead of being indexed, and the workspace dashboard says why. Nothing is charged for them.
- Search keeps working Everything already indexed stays searchable. Waiting files are tried again when the monthly units start over.
- Make room or move up Exclude folders you never need to search, or move to a larger plan for higher limits.
What Smart Search does with your files
What happens, who does it, and where the result is kept.
- Off by default
- A workspace has no index until its owner turns Smart Search on. Nothing is processed before that.
- Processing is automatic
- Extracted text, descriptions and vectors are stored only in your workspace's search index and its cache, in your workspace's region, and are deleted with the file or when you turn Smart Search off. Cloudflare Workers AI processes the files you opted in; it does not keep the inputs or train on them (Cloudflare Workers AI data-usage terms).
- The index stays in your region
- Each workspace's index lives in its own Durable Object, in the storage region chosen when the workspace was created.
- You choose what is covered
- Files in excluded folders are not indexed, and turning "Also index existing files" off limits it to files added from then on.
- Vault files are never indexed
- Vault Spaces are end-to-end encrypted: their files reach dosya as ciphertext, and an encrypted Space cannot hold an index.
- Deleting a file deletes its index data
- A file moved to Trash leaves the index at once. Its extracted text is kept only while it sits in Trash, so a restore does not spend units again, and is deleted when the file is.
- No faces
- There is no face recognition: nothing detects, groups or matches faces. Photo descriptions say what is in a scene.
- Only people with access see results
- Search answers inside one workspace, for members who can open its files, and drops anything in a locked or hidden folder they cannot see.
For files nobody else should process at all, use the Vault.
Questions
Which plans include Smart Search?
Every paid plan: Starter, Plus, Pro, Business. Free does not. During a 7-day card trial a workspace can index up to 200 photos and 100 documents. If a plan ends, indexing stops, and after 30 days Smart Search is switched off for that workspace.
Does it work in the Vault?
No, by design. Vault files are encrypted on your device before upload, so dosya only ever holds ciphertext and there is nothing to index. An encrypted Space cannot have Smart Search turned on.
Can members turn it on or change it?
The workspace owner can, and so can anyone whose role has the Change AI search permission. They can switch it on or off and choose the excluded folders. Indexing spends the owner's processing units.
What does it cost me?
Nothing on top of your plan, within its limits. A processing unit is one photo, or about 4,000 characters of document text (roughly 600 English words). Each plan has a one-time allowance for the files already there when you turn it on, and a monthly allowance for new ones. On a family plan the allowances are the payer's, shared by the family.
Which languages does it understand?
The bge-m3 model behind meaning search is multilingual, so a query and a document do not need to be in the same language to match. Exact-text matching treats the Turkish I, İ and ı as the same letter, and ignores spaces and dashes inside codes.
How do I turn it off?
Switch "Search inside files and photos" off in Settings → AI search. The workspace's index is wiped straight away, and every piece of extracted text and description is queued for deletion by a background job that runs every two minutes. Turning it on again starts a new index.
Are there limits on searching?
Each person can run up to 60 searches a minute. Documents over 25 MB are not indexed, and a document is indexed up to 300 pages, and up to about 800,000 characters of text; the rest of a longer one is left out.
Search what is inside, not only what it is called
Start on Free. Once your workspace is on a paid plan, turn Smart Search on from Settings → AI search; it searches inside your files after you turn it on and the files are processed by Cloudflare Workers AI.