Chat with over your 20 years of emails.

• Install Ollama and ensure ollama daemon is running.
• Python 3.13 with Poetry.

cd mymails
poetry install

# Pull models
ollama pull nomic-embed-text
ollama pull llama3.1

OfflineIMAP can be used to sync emails from your email provider to your local machine.

A sample offlineimap configuration syncing emails from iCloud to ~/Mail/iCloud.

file: ~/.offlineimaprc

[general]
accounts = iCloud
ui = ttyui

[Account iCloud]
localrepository = iCloudLocal
remoterepository = iCloudRemote

[Repository iCloudLocal]
type = Maildir
localfolders = ~/Mail/iCloud

[Repository iCloudRemote]
type = IMAP
remotehost = imap.mail.me.com
remoteuser = example@icloud.com
remotepass = some-secret-application-pass
ssl = yes
sslcacertfile = /etc/ssl/cert.pem

# run offlineimap
offlineimap

poetry run python -m mymails.ingest \
  --maildir ~/Mail/iCloud \
  --persist ./chroma \
  --max-messages 100 \
  --reset

The ingestion supports a progress bar, graceful interruption (Ctrl-C), and automatic resume:

• By default, a checkpoint file is written under the persist directory named like ingest_<collection>.state. Each processed message is recorded by a deterministic base id.
• On restart, ingestion skips already-processed messages and continues from where it left off.
• Use --no-progress to disable the tqdm progress bar.
• Use --no-resume to force a full reprocess (or --reset to wipe the DB first).
• Use --checkpoint /custom/path.state to choose a custom checkpoint location.
• Use --batch-size N to control embedding/upsert batch size (default 64).

poetry run python -m mymails.query \
  --persist ./chroma \
  --question "What were my travel plans June of 2024?"

Optional flags:

• --collection emails to change collection name
• --model llama3.1 to switch the chat model
• --k 5 top-k documents to retrieve
• --max-messages limit number of emails ingested for quick tests

On a M3 Macbook Pro, average ingestion time is 2 minutes for 1000 emails.