SahajMails

Personalized bulk email that runs on your machine.

Sahaj means simple and natural. So does this.

pip install sahajmails
sahajmails

That's it. A local web app opens. Load a spreadsheet, write your email, send.

Your contacts, your password and your drafts never leave your computer. There is no account, no server of ours, and no telemetry.

What it does

App Preview

Campaigns & Setup	Execution & Logs
AI Personalization	Plugins & Extensibility
Deliverability	Suppression List

The 60-second version

pip install sahajmails
sahajmails                 # opens http://localhost:8000

1. Settings → your address and an app password → Test connection
2. New campaign → drop your CSV → write your email → watch the live preview
3. Review → pre-flight + send yourself a test
4. Send

Prefer a terminal? Everything is a subcommand:

sahajmails init                                   # starter files here
sahajmails check contacts.csv -t email.md -s "Hi {{ first_name }}"
sahajmails send  contacts.csv -t email.md -s "Hi {{ first_name }}" --dry-run

Or import it:

from sahajmails import send

send(contacts="contacts.csv", template="email.md", subject="Hello {{ first_name }}")

Writing the email

Your spreadsheet columns become placeholders. Casing and spacing don't matter — {{ firstName }}, {{ first_name }} and {{ FIRST_NAME }} are the same column.

Hi {{ first_name | default("there") }},

{% if company %}I saw the news about {{ company }} — congratulations.{% endif %}

The rest of this email is identical for everyone.

Best,
Sahaj

default(...) matters: without it, a blank cell stops the send rather than quietly mailing "Hi ,". That is deliberate.

AI personalization

Wrap the one sentence you want personalized. Everything outside the block is byte-identical for every recipient — that is enforced by the design, not by asking the model nicely.

Hi {{ first_name }},

{% ai "opener" max_words=25 fallback="Hope your week is going well." %}
One warm sentence about {{ company }}. No greeting, no exclamation marks.
{% endai %}

I'm writing because …

• The model receives only that instruction and that contact's variables.
• Output is capped, stripped of markup, and checked against your banned phrases.
• Anything that fails becomes your fallback. A flaky API can never block a send.
• You review every line before anything is sent, and your edits are used verbatim.

Works with Claude, OpenAI, or any OpenAI-compatible endpoint — including Ollama and LM Studio running locally, so the "nothing leaves your machine" promise stays true.

Documentation

Security in one paragraph

The app binds 127.0.0.1 only. It is protected by a session token, a Host allowlist (which is what stops DNS-rebinding attacks that same-origin policy does not), CSRF headers, and a sandboxed template engine that escapes contact data so a spreadsheet can never inject markup into your mail. Passwords are kept in memory unless you explicitly ask to store them. Full detail in docs/security.md; report anything you find via SECURITY.md.

Contributing

git clone https://github.com/sahajrajmalla/sahajmails.git
cd sahajmails
make install     # needs pip >= 25.1
make check       # ruff, mypy --strict, pytest

See CONTRIBUTING.md. The whole test suite runs offline — there is a built-in SMTP stub and a fake AI backend, so you never need real credentials to develop.

License

MIT — free for personal and commercial use.