Sortio has two ways to organize a folder, and the most common new-user question is "which one do I use?" Both are AI-driven, both take a plain-English description of what you want, and both produce organized output. They make a different trade-off, and getting the choice wrong costs either time or money.
This post is the practical guide. What each mode is, how they behave on small folders vs. large drives, how the cost model differs, and a decision rubric for picking the right one. The short answer is at the top, the nuance is below.
The short version
AI Sort runs an LLM per file. Best for batches under ~5,000 files, ambiguous content, and one-off sorts. Rule Builder generates a deterministic rule once and runs it forever for free. Best for large drives, predictable patterns, and watch-folder automation. Most heavy users run both: AI Sort to discover the patterns, Rule Builder to run them at scale.
What AI Sort does
AI Sort takes a prompt like "sort my Downloads folder. Invoices go to ~/Documents/Finance, screenshots go to ~/Pictures/Screenshots, installers go to the Trash." Sortio iterates the folder, reads each file's content (PDF text, image OCR, document body), sends it to an LLM with the prompt as context, and the model returns a routing decision per file. Sortio shows the proposed moves in a preview, you confirm, and the sort runs.
The strength is intelligence per file. The model understands what an invoice looks like even if the filename does not say "invoice." It tells the difference between a chat screenshot and a receipt screenshot. It picks the right vendor name out of a noisy header. For folders where every file is different and the routing decision depends on reading the file, AI Sort is the right tool.
The cost is inference time and credits. Every file you sort consumes a small amount of credit on the managed AI tier, or some local inference time if you run Ollama. For a 100-file folder this is invisible. For a 50,000-file drive this is a real cost in both time and credits.
What Rule Builder does
Rule Builder takes the same kind of description ("move every PNG that starts with Screen Shot to ~/Pictures/Screenshots") and instead of running inference per file, it generates a deterministic rule. The rule is a small structured config (match by extension, by filename pattern, by metadata, by date) that Sortio executes directly without calling an LLM.
The strength is speed and cost. A rule sorts 50,000 files in seconds. There is no per-file inference, no token cost, no roundtrip to an AI provider. Once the rule exists, it runs the same way forever, even if you go offline. It is the right tool when the pattern is stable and you want it to run unattended.
The cost is up-front design and limited intelligence. Rules see file metadata and filename patterns. They do not read PDF content, they do not OCR images, they do not understand "is this a contract." If the pattern is more sophisticated than what metadata can express, a rule cannot do it; you need AI Sort.
The 5,000-file threshold
Sortio's honest guidance from a few thousand real-user sort runs: AI Sort works best on batches under about 5,000 files. Above that, three things happen that push the workflow toward Rule Builder.
- Time. 10,000-file AI sorts take 15+ minutes on cloud and longer locally. You can wait, but it is not the instant feedback that makes Sortio useful.
- Cost. Free-tier credits run out fast at 10,000+ files. Pro's 5,000 monthly credits cover most users but not the data-hoarder profile.
- Repetition. Large drives usually have a long tail of similar files (a folder of 30,000 photos, a folder of 50,000 documents). Running AI per file when the routing decision is the same for every file is wasteful.
Above the threshold the right workflow is: run AI Sort on a representative sample (a few hundred files), see what patterns the model is using, promote those patterns to rules, then let the rules sweep the rest of the drive. This is the same idea as sample-and-script in data work: use the expensive intelligent tool to find the structure, then apply a cheap deterministic tool at scale.
Three real workflows
Workflow 1: Downloads folder cleanup (AI Sort)
Folder size: 1,500 files. Content: invoices, receipts, installers, ZIPs, miscellaneous downloads. The routing decisions need content reading (an invoice from a new vendor has no filename pattern; an old ZIP is not obviously stale from metadata alone). AI Sort is the right tool. One prompt, one run, done in two minutes. See the full walkthrough in our piece on keeping your Mac Downloads folder clean.
Workflow 2: 30,000-photo migration (Rule Builder)
Folder size: 30,000 photos exported from Lightroom, all with consistent EXIF metadata. The routing decision is "by date, then by camera." This is a pure metadata sort: Rule Builder generates a deterministic rule that reads EXIF and routes by year/month/ camera in seconds. AI Sort would work too but would be slow and wasteful. Rule Builder is the right tool.
Workflow 3: 8,000 mixed legal documents (both)
Folder size: 8,000 PDFs from a small-firm legal practice. About 70 percent are predictable (deposition transcripts named by convention, court filings with a consistent docket pattern). 30 percent are not (random vendor invoices, scanned client documents, misc correspondence). The right workflow is two-pass: Rule Builder handles the 70 percent that have a stable pattern, AI Sort handles the 30 percent that do not. Both run in the same sort, rules first.
A decision rubric
Pick AI Sort when at least one of these is true:
- The folder has fewer than 5,000 files.
- The routing decision needs the file's content (you cannot decide by filename and metadata alone).
- This is a one-off sort you will not repeat.
- You do not yet know the patterns in your data.
Pick Rule Builder when at least one of these is true:
- The folder has 5,000+ files.
- The routing decision is the same every time and depends on metadata or filename pattern.
- You want this to run unattended (watch folder, scheduled run).
- You want to minimize credit consumption.
Run both when:
- The folder is mixed: some files have stable patterns, others do not. Rules handle the easy cases, AI Sort handles the long tail.
- You are migrating a large archive: AI Sort to discover patterns on a sample, Rule Builder to execute at scale.
Cost considerations
Sortio Free includes 10 AI sort credits a month and rule-based sorting up to 5 rules. Pro at $14.99/month or $99/year gives 5,000 AI credits a month and unlimited rules, plus watch folders and scheduled runs. For most users the credit allowance is plentiful; the move to Pro is usually driven by the watch-folder and scheduled-run features, not by hitting the credit cap.
If you run high-volume AI sorts and want predictable cost, the Ollama local mode is free of per-sort cost (electricity aside). See our piece on local AI vs cloud AI for the speed and accuracy trade-off.
Promoting an AI Sort to a rule
The most efficient long-term workflow is to start with AI Sort, then promote successful patterns to rules. After a sort run, open the activity log, right-click any decision, and choose "Make this a rule." Sortio passes the natural-language description to the Rule Builder, which produces a deterministic rule and shows you a preview against the same files. If the preview matches what AI Sort did, save the rule and it runs on future sorts without calling an LLM.
For users coming from Hazel, this is the bridge: you describe the rule the way you described it to Hazel, and Sortio builds the structured config under the hood. The Hazel-to-Sortio migration guide walks through the common Hazel patterns and their Rule Builder equivalents.
FAQ
What is the difference between AI Sort and AI Rule Builder in Sortio?
AI Sort takes a plain-English prompt and uses an LLM to make a per-file routing decision. Every file gets read by the model and routed individually. AI Rule Builder takes a plain-English description and generates a deterministic rule (extension, filename pattern, metadata) that runs without an LLM. AI Sort is best for small batches and ambiguous content; Rule Builder is best for large drives and predictable patterns.
When does AI Sort stop working well?
AI Sort works best on batches under about 5,000 files. Above that, inference time gets long enough that you would rather wait for a deterministic rule to run, and credit consumption gets expensive on the managed cloud tier. The recommended workflow for large drives is to run AI Sort on a representative sample to figure out the right rules, then promote them to AI Rule Builder, then run the rules over the full drive.
Does AI Rule Builder use credits?
Rule generation uses one credit (the AI describes-to-rule step). Rule execution does not: once the rule exists, it runs deterministically on your machine for free, no matter how many files. This is why Rule Builder is the right tool for large drives: you pay for the AI step once and the rule handles the volume.
Can I mix AI Sort and rules in the same sort run?
Yes. Sortio applies rules first (deterministic, fast) and uses AI Sort to handle anything the rules did not match. This is the most efficient workflow for mixed folders: rules catch 80 percent of the predictable cases, AI handles the long tail. You can also stack the rules: first rule routes screenshots, second routes installers, third routes invoices, AI catches whatever did not match.
When is AI Sort clearly the right tool?
Three situations. First, ambiguous content where the routing decision needs to read the file (an invoice that does not say "invoice" in the filename, a contract that could be filed under two counterparties). Second, small batches where you only run the sort once or twice and a rule is not worth building. Third, exploratory sorts where you do not yet know the patterns in your data: AI Sort surfaces them so you can decide what rules to build.
When is Rule Builder clearly the right tool?
Three situations. Large drives (10,000+ files) where AI inference would be slow and expensive. Stable patterns where the routing decision is the same every time (Screenshots go to ~/Pictures/Screenshots, period). Recurring workflows where a rule needs to run unattended for years (watch folders for a project, monthly archive sweeps).
How do I promote an AI Sort into a rule?
After an AI Sort run, the right-click menu on any routing decision lets you "Make this a rule." Sortio passes the prompt context to the Rule Builder, which converts the natural-language description into a deterministic rule and lets you preview it. If the preview matches what AI Sort did, you save the rule and it runs on future sorts without consuming AI credits.