mu4e-llm

AI-powered email assistance for mu4e using LLM providers via llm.el.

%%{init: {'theme': 'base', 'themeVariables': { 'primaryColor': '#4a6fa5', 'primaryTextColor': '#fff', 'primaryBorderColor': '#2d4a6f', 'lineColor': '#5c7caa', 'secondaryColor': '#e8f0fe', 'tertiaryColor': '#f5f5f5'}}}%%
flowchart LR
    subgraph input [" "]
        direction TB
        E["📧 Email Thread"]
    end

    subgraph features ["mu4e-llm"]
        direction TB
        S["📋 Summarize"]
        R["✍️ Smart Reply"]
        T["🌐 Translate"]
    end

    subgraph output [" "]
        direction TB
        O1["Summary"]
        O2["Draft"]
        O3["Translation"]
    end

    E --> S --> O1
    E --> R --> O2
    E --> T --> O3

    style input fill:#f9f9f9,stroke:#ddd
    style features fill:#4a6fa5,stroke:#2d4a6f,color:#fff
    style output fill:#f0f7ff,stroke:#4a6fa5

Loading

Features

• Thread Summarization: Get detailed or executive summaries of email threads
• Smart Reply Drafting: Generate context-aware email replies with iterative refinement
• Translation: Translate messages, threads, or selected text to multiple languages
• Provider Agnostic: Works with OpenAI, Claude, Gemini, Ollama, and any llm.el provider
• org-msg Integration: Drafts use org-mode syntax for styled HTML emails

Requirements

• Emacs 28.1+
• mu4e (email client)
• llm.el 0.17+ (LLM provider abstraction)
• An LLM provider (OpenAI, Claude, Gemini, Ollama, etc.)

Installation

Using straight.el

(straight-use-package
 '(mu4e-llm :type git :host github :repo "sillyfellow/mu4e-llm"))

Manual Installation

1. Clone or download this repository to ~/.emacs.d/mu4e-llm/
2. Add to your init.el:

(add-to-list 'load-path "~/.emacs.d/mu4e-llm")
(require 'mu4e-llm)
(with-eval-after-load 'mu4e
  (mu4e-llm-setup))

Quick Start

1. Configure an LLM provider (see Provider Setup)
2. Open mu4e and navigate to an email
3. Press C-c a e s to summarize the thread
4. Press C-c a e r to generate a smart reply

Commands

Main Commands (C-c a e prefix)

Key	Command	Description
s	mu4e-llm-summarize	Generate detailed thread summary
S	mu4e-llm-summarize-executive	Generate brief executive summary
r	mu4e-llm-draft-reply	Generate smart reply draft
R	mu4e-llm-draft-refine	Refine current draft
n	mu4e-llm-draft-compose	Compose new email with AI
t	mu4e-llm-translate-message	Translate current message
T	mu4e-llm-translate-thread	Translate entire thread
a	mu4e-llm-abort	Abort current operation
?	mu4e-llm-help	Show help

Draft Buffer Keys

Key	Command	Description
C-c C-r	Refine	Refine with custom instruction
C-c C-s	Shorten	Make draft more concise
C-c C-p	Polite	Make draft more polite/professional
C-c C-f	Finalize	Accept draft and open in compose
C-c C-k	Cancel	Discard draft

Workflow Diagrams

Smart Reply Workflow

%%{init: {'theme': 'base', 'themeVariables': { 'primaryColor': '#4a6fa5', 'primaryTextColor': '#fff', 'lineColor': '#5c7caa'}}}%%
flowchart TD
    A["📧 View email"] -->|"C-c a e r"| B["🤖 Generate draft"]
    B --> C["📝 Review in *draft* buffer"]

    C --> D{Done?}

    D -->|No| E["🔄 Refine"]
    E -->|"C-c C-r/s/p"| B

    D -->|Yes| F{Accept?}
    F -->|"C-c C-f"| G["✉️ Open in compose"]
    F -->|"C-c C-k"| H["🗑️ Discard"]

    style A fill:#f0f7ff,stroke:#4a6fa5
    style B fill:#4a6fa5,stroke:#2d4a6f,color:#fff
    style C fill:#e8f4e8,stroke:#4a8f4a
    style G fill:#4a8f4a,stroke:#2d6f2d,color:#fff
    style H fill:#f5f5f5,stroke:#999

Loading

Architecture

%%{init: {'theme': 'base', 'themeVariables': { 'primaryColor': '#4a6fa5'}}}%%
flowchart TB
    subgraph emacs ["Emacs"]
        MU4E["mu4e"] --> |email data| CORE
        subgraph pkg ["mu4e-llm"]
            CORE["core"] --> SUM["summary"]
            CORE --> DFT["draft"]
            CORE --> TRN["translate"]
        end
    end

    subgraph external ["External"]
        LLM["llm.el"] --> API["OpenAI / Claude / Ollama"]
    end

    CORE <--> LLM

    style emacs fill:#f9f9f9,stroke:#ddd
    style pkg fill:#e8f0fe,stroke:#4a6fa5
    style external fill:#fff5e6,stroke:#d4a84b

Loading

Configuration

All options are customizable via M-x customize-group RET mu4e-llm RET.

LLM Provider Settings

;; Use a specific provider (overrides fallback)
(setq mu4e-llm-provider (make-llm-openai :key "your-api-key"))

;; Or leave nil and use fallback variable (see Provider Setup)
(setq mu4e-llm-provider nil)
(setq mu4e-llm-provider-fallback-variable 'my/llm-provider)

;; Response creativity (0.0 = focused, 1.0 = creative)
(setq mu4e-llm-temperature 0.7)

;; Maximum response length
(setq mu4e-llm-max-tokens 2048)

Thread Extraction

;; Maximum messages to include in thread context
(setq mu4e-llm-max-thread-messages 20)

;; Maximum characters per message body
(setq mu4e-llm-max-message-length 4000)

Caching

;; Enable/disable summary caching
(setq mu4e-llm-cache-summaries t)

;; Cache expiry in seconds (default: 1 hour)
(setq mu4e-llm-cache-ttl 3600)

;; Clear cache manually
(mu4e-llm-clear-cache)

Draft Settings

;; Default persona style: professional, friendly, formal, concise
(setq mu4e-llm-draft-persona 'professional)

;; Include thread summary in draft buffer
(setq mu4e-llm-draft-include-summary t)

Translation

;; Available languages (alist of display-name . code)
(setq mu4e-llm-languages
      '(("English" . "en")
        ("German" . "de")
        ("French" . "fr")
        ("Spanish" . "es")
        ("Japanese" . "ja")
        ("Chinese" . "zh")))

;; Default target language
(setq mu4e-llm-default-target-language "en")

Prompt Customization

All LLM prompts are customizable via M-x customize-group RET mu4e-llm RET:

Variable	Purpose
mu4e-llm-summary-standard-prompt	Standard thread summary
mu4e-llm-summary-executive-prompt	Executive summary
mu4e-llm-draft-reply-prompt	Reply drafting
mu4e-llm-draft-refine-prompt	Draft refinement
mu4e-llm-draft-compose-prompt	New email composition
mu4e-llm-draft-persona-descriptions	Persona style definitions
mu4e-llm-translate-message-prompt	Single message translation
mu4e-llm-translate-thread-prompt	Thread translation
mu4e-llm-translate-text-prompt	Text/region translation

Example customization:

;; More detailed summaries
(setq mu4e-llm-summary-standard-prompt
      "Provide an extremely detailed summary of this email thread.
Include every decision, action item, and participant opinion.

EMAIL THREAD:
%s")

;; Custom persona
(add-to-list 'mu4e-llm-draft-persona-descriptions
             '(casual . "Write casually, like texting a friend."))

Provider Setup

Using a Fallback Variable (Recommended)

If you have a global LLM provider variable, point mu4e-llm to it:

;; If you have a global provider variable like my/llm-provider:
(setq mu4e-llm-provider-fallback-variable 'my/llm-provider)

;; Now changing my/llm-provider automatically affects mu4e-llm

Direct Provider Configuration

;; OpenAI
(setq mu4e-llm-provider
      (make-llm-openai :key (auth-source-pick-first-password
                              :host "api.openai.com")))

;; Claude
(setq mu4e-llm-provider
      (make-llm-claude :key (auth-source-pick-first-password
                              :host "api.anthropic.com")))

;; Ollama (local)
(setq mu4e-llm-provider
      (make-llm-ollama :chat-model "llama3.1"))

org-msg Integration

mu4e-llm generates drafts in org-mode syntax. If you use org-msg, your drafts will automatically render as styled HTML emails.

Without org-msg, drafts are plain text with org-mode formatting that you can copy to your compose buffer.

Troubleshooting

"No LLM provider configured"

Either:

1. Set mu4e-llm-provider to an llm.el provider object
2. Set mu4e-llm-provider-fallback-variable to point to your global provider variable

Summaries are stale

Clear the cache with M-x mu4e-llm-clear-cache or C-u M-x mu4e-llm-summarize.

Thread extraction incomplete

Try increasing mu4e-llm-max-thread-messages or ensure your mu index is up to date with mu index.

License

MIT. See LICENSE.

Author

Dr. Sandeep Sadanandan sillyfellow@whybenormal.org