Olive Mail

Open-source · Local CLI · Mac · Linux

Email as Context for Agents
Controlled, Private, Token-Efficient

Existing mail accounts hold crucial context for real work
Agents need that context. They don't need your passwords, unrestricted right to send, or sensitive parts of emails.

Olive Mail puts you in control
It's a local mediator that you authenticate. Agents never touch credentials. They use Olive Mail to search, read, draft or send.

Olive Mail is more than a security layer
It also aims to offer granular permission control, fast cached access, semantic search, and custom redaction plugins.

Get Started

Run This to Install

agent-user 
$ brew install nohype-ai/tap/olive-mail
↓ Copy
  1. 1

    Run the above command

    It will use Homebrew to install olive-mail, either from the prebuilt binary or by building it with Swift.

  2. 2

    Add an email account

    olive-mail account add you@work.com adds the credentials for that email account.

  3. 3

    Tell your agents

    Point your agents to the olive-mail CLI so they know to use it. An explicit agent skill is also coming.

Why We Built This

Common Practices
Are Often Symptoms – Not Solutions

Agents wielding email can feel like a superpower. But the way we connect both is still immature: either directly, via dedicated connector, or giving the agent its own email account. None of those approaches fit real work.

Locked Out

Typical agents see no real email context as they get locked into their own little mailbox. They miss the threads that actually matter: clients, legal, finance – the work.

All-Powerful

Or agents can do it all – holding the keys to your kingdom. One prompt-injected message, and the inbox is public. They can speak as you, email anyone, destroy anything.

Leaking Sensitive Data

Real emails often contain sensitive data that a business needs to filter or redact before letting agents read emails at all.

Slow and Expensive

Agents accessing mail accounts as such is slow, loads redundant data, offers no meaningful search, and costs a lot of tokens.

Where We're Going

A Security and Context Gateway – Not Another AI Inbox

  • Protected Credentials

    Your credentials are isolated from agents, by actual account separation. You authenticate. They call olive-mail with a scoped interface, not the password file. You monitor and revoke.

  • Granular Permission Control

    By unix account and mail account. Reading permissions by mailbox, email field, sender. Permissions to draft and send, by destination. Rate limits. Whatever it takes.

  • Redaction of Sensitive Data

    Sensitive data – PII, legal, finance – stripped locally by custom plugins, so a remote model never sees what it should not. The mailbox stays the system of record. The agent sees a view.

  • Fast and Cheap Access

    A local cache across agents for super fast responses. Get locally generated summaries. Extract content from HTML. Only pay for (remote-) tokens where frontier AI is actually needed.

  • Outbound Authenticity

    Agents can but don't have to act as spokesperson for you or your company. You can employ them like interns, so they are in the loop while you're still the one who speaks.

  • Semantic Retrieval

    Find meaning that matches a prompt – not just letters that match a search term. Retrieve relevant content across accounts and mailboxes instead of stuffing them into the agent.

Like the Idea?

★ Star Olive Mail on GitHub