MBox Trainer helps you import real email conversations, process them with local or remote LLMs, review useful candidates, and prepare clean JSONL training examples — so you can see what your own archive can become before choosing a paid plan.
Support threads, project updates, sales replies, operations messages, and internal discussions often contain real examples of how your organization communicates. The hard part is turning that messy archive into clean, usable model data.
Threads include quotes, signatures, duplicates, side chatter, private details, and half-finished conversations.
A model cannot learn much from a raw inbox dump. It needs focused prompts, useful completions, and consistent output structure.
The goal is not to blindly automate training data. The goal is to help you find better candidates faster and keep control before export.
MBox Trainer gives the workflow a clear path: import the source data, process it with the right profile, review the candidates, and export the examples that are ready to use.
Start with MBOX files, mailbox sources, or manual import paths.
Use processing profiles and prompt templates to guide what the app should look for.
Classify, clean, extract, naturalize, score, and prepare candidate examples.
Approve, reject, edit, and understand why each candidate was suggested.
Generate local training-data files that are ready for your fine-tuning workflow.
The app is designed around the work between “I have old email” and “I have a dataset worth training on.”
Create reusable workflows for the kinds of examples you want to find, clean, score, and export.
Tune the instructions used for classification, extraction, privacy cleanup, quality checks, and naturalization.
Prepare data more responsibly by cleaning sensitive details before examples move closer to export.
Keep the human in the loop with approval, rejection, editing, and evidence for why a candidate was created.
Use the desktop app with local or remote model endpoints depending on your workflow and comfort level.
Export clean training examples in a practical format for downstream fine-tuning experiments.
Start with a 30-day Basic workflow trial, then choose the yearly plan that fits your archive size, processing volume, and export workflow.
For solo builders validating smaller email-to-dataset workflows.
$120/year
For regular dataset-building work across multiple projects.
$240/year
For larger archives, heavier processing, and faster sync workflows.
$590/year
Yearly pricing shown. Monthly billing options may be available at checkout. The Basic workflow trial lets you test importing, processing, and reviewing candidate examples before paying.
Exporting finalized datasets requires an active paid plan. Remote model usage may require separate provider/API costs depending on the endpoint you configure.
Download the desktop app, run the Basic workflow trial, and see whether your old conversations can become useful training examples before choosing a paid plan.
You can import, process, review, and inspect candidate examples during the trial. Exporting finalized datasets requires an active paid subscription.