Workflow guide

Reading RFQ emails into line items

Short answer: Intake is two folders: request files in, draft files out. Staffbox reads the plain-text body and holds no mail password.

Before any price is worked out, a request email has to reach the worker. For the IT provider setting it up, mailbox access is usually the riskiest part. Staffbox needs none: intake is a folder, and pricing is covered on the RFQ automation page. Send up to three requests below; drafts come back within 2 business days.

Updated 13 October 2026

Intake is a folder, so there is no mail login to configure

One request in, three things out. Nothing in the path holds a mail password or sends. Source: staffbox/inbox.py and README, github.com/Staffbox-ai/staffbox, read 13 Oct 2026.

Intake is a folder. Staffbox reads request files from one folder and writes drafts to another. There is no mail login to configure. A typical layout:

requests/          new request files, .eml or .txt
requests/done/     files already processed (moved here by the worker)
drafts/            one draft .eml per request, same file name
vault/log.md       one line per draft, append-only

The command, as printed by staffbox --help:

staffbox inbox VAULT --in requests/ --drafts drafts/ [--model qwen3:8b]

Optional flags set the model host and the context size. The worker takes the files in file-name order and handles one request at a time. Concurrent requests queue, so a burst of forty emails is a queue of forty, not forty at once.

Only the plain-text body is read; quoted history and attachments are not

Input	What the worker uses	Note
`.eml`	The From and Subject headers and the plain-text part of the body.	An email with no plain-text part is read as an empty body. HTML-only senders need a plain-text copy made at the sync step.
`.txt`	The whole file is the request text. The file name, with dashes turned into spaces, becomes the subject. There is no sender, so the greeting is generic.	Fine for a pasted request or a test.
Quoted history	Lines starting with > are removed before the request is read.	Forwarded blocks that are ordinary text, not quoted with >, stay in the request.
Attachments	Not read.	PDFs, spreadsheets, images and drawings are outside what we have tested. We make no claim about them.

These details come from reading the mailbox module in the open repository on 13 October 2026, not from a marketing description.

Each request yields one draft, one moved file and one log line

The request text goes to the same question-answering step used in the scorecard, with the company brain and the quote action.
The result is written to drafts/ as an .eml file with the same name as the request. The subject gets "Re:" if it does not already start with it.
The request file moves to requests/done/, so a re-run never drafts the same request twice.
One line is appended to the log: the request file name, the draft file name, the first part of the answer, and the words "not sent: a person sends".

The draft itself is described on draft replies that cite the price-list row.

Try the intake yourself

The repository includes a three-email sample inbox from the fictional Fieldstone IT, with the commands to run it:

cp -R examples/fieldstone-it/inbox-sample /tmp/requests
cp -R examples/fieldstone-it/vault /tmp/fieldstone-vault
bin/staffbox inbox /tmp/fieldstone-vault --in /tmp/requests --drafts /tmp/drafts --model qwen3:8b

The three emails show the cases intake has to cope with. One asks for 15 laptops and carries a quoted older message below the request. One asks for MacBook Pros, which are not on the Fieldstone list, and ends with a phone signature. One asks for a single firewall installed. The model must be available through Ollama on the machine.

Getting mail into the folder is per-site work, with stated limits

Staffbox contains no mail connector and no credentials. The folders can be filled by whatever your mail setup already offers: a mail client rule that saves messages to a folder, a sync tool, or an export. If the folders are synced with your mail system, the mail stays where that system already keeps it, and Staffbox adds no new copy outside the box. That wording is from the data policy in the repository.

Which of those routes suits you depends on your mail system. It is set up at install by your IT provider and is per-site work. We have not published a recording of an end-to-end run against a live Microsoft 365 or Google Workspace mailbox. If you need one, ask for it in the free test (not measured yet).

Forwarding is the simplest habit. A staff member forwards a request to a dedicated quotes address, and the sync step turns that mailbox into the requests folder (our view).

Limits of intake today

Plain-text parts only. Attachments and HTML-only emails are not read.
One folder in, one folder out. Routing to different queues is a separate step; see email triage.
A person sends. No intake setting turns on sending; the code has none. See human approval.
No customer mailbox has run through it yet.

Questions

Can an AI read RFQ emails and extract line items?

Yes for the text of the email. Staffbox reads the plain-text body of each request file, finds the items and quantities, prices them from your list and writes a draft. It does not read attachments such as PDFs or drawings.

Does Staffbox need my email password?

No. It reads request files from one folder and writes drafts to another, and holds no mail credentials. How mail reaches the folder is set up at install by your IT provider.

What happens to a request after it is processed?

The request file moves to a done folder, a draft with the same file name is written to the drafts folder, and one line is added to the log.

Sources

Mailbox door behaviour: staffbox/inbox.py, README and docs/data-policy.md in github.com/Staffbox-ai/staffbox, read 13 Oct 2026.
Sample inbox: examples/fieldstone-it/inbox-sample in the same repository (fictional company).

Or call us: 404 493 6248 · How the free test works