Routing rules

A routing rule is what makes Scribe automatic. It says: when a file matching this name pattern appears in this input folder, run this workflow, on this model, using these reference documents, and write the result to this output folder.

Rules live in the Project's Scribe Config tab under Routing rules. A Project can have as many as it needs, for example one rule per document type or per supplier.

Before you start

You need an attached folder (Connect storage), at least one workflow enabled on the Project, at least one reference document, and the project.manage_scribe permission. If the Project has no workflows enabled, the builder is locked and tells you so.

Create a rule

Click New rule and fill in the form. The workflow you pick decides which fields appear.

Field	What to enter
Name	A label so the rule is recognizable later, such as "CoA files → Summary drafts".
Workflow	One of the workflows enabled on this Project.
Model	The AI model the rule runs on: a PharmFindr-hosted model or one of your own connected models. If the list is empty, connect a provider or check model access under Manage > AI Providers.
Reference documents	One picker per role the workflow needs, each with a short explanation. For example Certificate of Analysis Review asks for a Material Specification and a Reconciliation Report Template.
Filename pattern	Which files in the input folder trigger the rule. Defaults to * (every file).
Input folder	Where source documents arrive. Click Use as input on a folder in the browser.
Output folder	Where Scribe writes results. Click Use as output on a folder in the browser.

Then click Create rule.

Filename patterns

Patterns use * as a wildcard and ignore case. CoA-*.pdf matches CoA-Batch001.pdf and coa-batch002.PDF; * matches everything. Not sure what pattern to write? Click Not sure about the pattern? Build one from example filenames, paste two or more real filenames, and Scribe suggests one.

Source files must be PDF or Word (.docx). Scanned documents are not supported, because there is no OCR step.

Choosing folders

The folder browser shows the folders attached to the Project. You can pick the attached folder itself or any subfolder inside it. A rule watches its input folder and everything beneath it.

Keep input and output apart

Use different folders for input and output. Scribe ignores its own output, but separate folders keep your source documents and generated reports easy to tell apart and let you give reviewers access to just the reports.

If no folders are attached yet, the builder links you to Connections to attach one.

Reference documents and output type

The output format follows the output template you pick: a Word template produces a Word document and a PDF template produces a PDF. There is nothing separate to choose, and the builder warns you if the template's file type doesn't suit the workflow.

What triggers a run

Scribe checks the attached folders continuously. Every time a file is added or changed, it looks for rules that match, and each match creates one run. A run needs all of these to be true:

• The file is a PDF or Word document, and its name matches the rule's filename pattern.
• The file is inside the rule's input folder, or beneath it.
• The rule's workflow is still enabled on the Project.
• Scribe is enabled for your organization.
• The input folder is still attached, and its path still exists in Dropbox.
• The file isn't itself one of your reference documents. Dropping a template into an input folder by accident is detected and skipped.
• This exact file revision hasn't already been processed by this rule.

If any of these fails, nothing happens and no run appears. That is by design, so a stray file doesn't create noise — but it means a missing run is a sign to check this list.

Several rules, one file

If a file matches more than one rule, it creates one run per rule. This is how you get, for example, both a Word and a PDF version of the same result from different rules covering the same folder. Two rules with identical settings aren't allowed.

Editing a file again

For Summary Generation, saving a new version of a source file updates the existing output file in place instead of creating a second one, preserving values you've already edited by hand. Certificate of Analysis Review and Batch Record Review treat every file as new work.

Scribe never processes its own output

Generated files are recognized by their location and ignored if they appear in a watched folder, so an output folder inside an input folder won't cause a loop.

Manage and delete rules

Each rule appears as a card showing its name, input → output path, workflow, and model. Click a card to see a diagram of the flow. Rules can't be edited: to change a setting, such as the model or a reference document, create a new rule and delete the old one with the trash icon.

Status badges on a card flag problems:

Badge	Meaning
Workflow inactive	The rule's workflow was turned off. New files won't trigger runs until an Owner re-enables it.
Input path lost	The input folder no longer exists in Dropbox (deleted or renamed). Recreate it with the same path and it clears itself the next time a file is seen.
Output path lost	The output folder was deleted or renamed. Recreate it and the next run clears the flag. Scribe never recreates it for you.

Next step

See what a run does and how to review it: Runs and review.