TabMail
AI triage, summaries and drafts for the inbox you already have, in open clients
Emerged finds new products worth a look, daily — what’s new today
S
Sinabis Synthetic Email Dataset
A fictional company's decade of mail, invites and attachments, free to train on
Anyone building email search, e-discovery or inbox agents finally has a licence-clean, attachment-rich corpus far bigger and newer than Enron.
Sinabis GmbH has published about 10.1 million fictional corporate emails as .eml files, with 1.4 million .ics calendar invites and over 2 million PDF, DOCX, XLSX and image attachments, spanning December 2015 to July 2026. Every person and company is invented. Text came mostly from Gemma 4 and Qwen models, with structure, timestamps and some invoices built by a rule-based generator. English makes up 87%; 28 other languages share the rest. CC BY 4.0, roughly 187 GB unpacked, on Hugging Face.
More like this
Pidgeon
A shared mailbox where every agent gets its own address
Clova
Give each company its own address; anyone else who writes to it is bounced
Egma
Open-source simulation testing and monitoring for voice agents
DataAnonymiser
Masks personal data before it reaches an AI chat, then puts it back in the reply
lieflat-less-ai-tone
An agent skill that strips AI tone from Chinese text, built on a corpus study