MailDesk docs
Get MailDesk
Pro

MailDesk AI — features

This page describes each AI action in MailDesk Pro: what it does, how you reach it, and what it is given to work with. Every action runs only on the email or thread you are working on, and only when AI is enabled for that mailbox. For the full data boundary, see What MailDesk AI can and cannot access.

6 min read Pro

Available in: MailDesk Pro. MailDesk Basic has no AI.

This page describes each AI action in MailDesk Pro: what it does, how you reach it, and what it is given to work with. Every action runs only on the email or thread you are working on, and only when AI is enabled for that mailbox. For the full data boundary, see What MailDesk AI can and cannot access.

A reminder that applies to every feature here: AI output is a suggestion, not a verdict. Read a summary, a draft, or a security result with normal care before you act on it.

Security scan (phishing & impersonation check)

When you open an email, MailDesk can give you a structured second opinion on whether the message looks like phishing, impersonation, or another social-engineering attempt — before you click any links.

An email opened in Safe Mode: MailDesk has blocked suspicious content and offers Check with AI, Show Once, and Trust Sender

What it checks

  • Sender authenticity — display name versus the actual address, look-alike domains, unusual reply-to fields.
  • Phishing indicators — urgency and pressure language, credential or payment prompts, redirection language.
  • Suspicious link patterns — URL shorteners, look-alike (homograph) domains, link text that does not match where the link actually goes.
  • Impersonation and intent — CEO or vendor impersonation and business-email-compromise patterns.

How to read the result

The result is a structured assessment with a risk level:

Risk level What it means Suggested action
Low No notable indicators Treat as normal
Medium Some suspicious patterns Verify with the sender out of band before acting
High Strong indicators of an attack Do not click links, do not reply, alert your administrator

The scan reads the sender address and the message text only; it does not open attachments. It is tuned to keep false positives low, so ordinary marketing mail, newsletters, and order confirmations are not flagged. A Low result is reassurance, not a guarantee — always verify a suspicious sender through a separate channel.

When the scan finds higher risk, MailDesk alerts you as you open the email (and, where the administrator has enabled the cautious "safe mode", shows extra confirm-before-you-trust controls on the message). A Low result is shown quietly.

The result is a signal, not proof. A High result means be careful, not certainly malicious; a Low result does not make an email safe. If a legitimate email is flagged High, tell your administrator so it can be reviewed.

Summarize a thread

When you open a long thread — many messages, several participants — Summarize produces a short recap so you can catch up at a glance.

The MailDesk AI panel showing an AI-generated thread summary with a short recap, an Action Items list, and Open Questions

How to use it

  1. Open the thread in MailDesk.
  2. Open the AI panel and click Summarize.
  3. MailDesk sends the thread to your configured AI provider.
  4. A summary appears in the panel, usually within a few seconds.

A summary typically covers the main decisions and commitments, open questions, key dates, and who said what. Only the messages in that thread are used — attachments, other threads, and your Odoo records are not included.

Ask AI

Ask AI is a multi-turn assistant that stays grounded in the thread you opened. Instead of a single button, it is a short conversation: you ask, it answers, and follow-up questions keep the same thread as context.

The MailDesk AI panel answering the question "What is this customer asking, and what should I do next?" with an AI-generated reply grounded in the open support thread

How to use it

  1. Open an email and click Ask AI (in the message toolbar, or from the message menu).
  2. The assistant starts grounded in that thread.
  3. Ask follow-ups in plain language — for example "What are the open questions?", "Who is involved in this thread?", "Help me write a reply in German", or "Make it shorter".
  4. Each answer can be opened straight into the MailDesk composer as a Reply, Reply All, or Forward, with recipients, subject, and threading prepared correctly.

Ask AI only ever sees that thread — your question, the conversation so far, and that thread's text. It does not reach into other mailboxes, other threads, or your CRM, calendar, or other Odoo data.

Draft a reply

Draft a reply proposes a reply you can review and edit before sending. The proposed text is grounded in the thread, so it reflects what was asked and what was promised.

The MailDesk AI panel showing an AI-generated reply body for a service-contract renewal, with Use as reply, Insert, and Copy actions and tone-refinement chips

How to use it

  1. Open the email in MailDesk.
  2. Open the AI panel and click Draft a reply (or ask for a reply from Ask AI).
  3. Optionally refine it with one click — for example Shorter, More formal, or Friendlier — or regenerate it.
  4. The draft opens in the composer; review it carefully, edit, and send.

The reply is generated in your active Odoo interface language by default, and you can ask for a different language. It produces the reply body only — no subject line, salutation header, or signature; you add those as usual.

The draft is built from the thread (each message's sender, recipients, date, subject, and text). It does not see attachments, other emails in your mailbox, or any Odoo data outside the thread — so always read the draft before sending; the AI can be wrong about facts or miss a subtle commitment.

Attachment Q&A (opt-in)

You can ask the AI about an attachment on a message — but only when an administrator has deliberately allowed it for that mailbox and you explicitly request it on that message.

How it works

  1. The mailbox must have Allow AI Attachment Analysis switched on (an administrator does this; it is off by default).
  2. On a message that has an attachment, choose to analyze the attachment.
  3. The AI returns what it found from the attachment.

Attachments are never sent automatically. With the switch off, you are told the feature needs to be enabled for the mailbox; nothing is sent. Some file types may not be supported, in which case MailDesk tells you so rather than sending anything.

Off by default, for good reason. Until an administrator opts a mailbox in, no attachment content is ever sent to the AI. This keeps documents inside Odoo unless you make a deliberate choice. See the data boundary in What MailDesk AI can and cannot access.

How access is controlled

Every AI action runs only when all of these allow it. They are checked on the server, not just hidden in the interface:

  1. Global switch — an administrator can turn off all AI at once.
  2. Per-feature switch — individual features can be turned off.
  3. Per-mailbox switch — Allow AI Features — off for a mailbox means no AI for that mailbox.
  4. A configured provider — a cloud provider with a valid key, or a local server.

You only ever see AI results for mailboxes you already have access to; the standard Odoo access rules still apply, and there is no special elevation.

Choosing and setting up a provider

You can use OpenAI, Google Gemini, Anthropic Claude, xAI Grok, DeepSeek, or a local / self-hosted server (for example Ollama, LM Studio, or vLLM). Administrators can also route different actions to different providers — for example a careful provider for the security scan and a fast one for summaries. To set this up, and to see the full list of what is and is not sent to a provider, follow What MailDesk AI can and cannot access → Turn AI on.

Available in MailDesk Pro 18.0.4.1.0+. Behaviour verified against the Pro AI service code and KB 257/259/260/261.