What Are AI Agents? [Updated Guide for 2026]

Gmelius Blog AI Assistants What Are AI Agents? [Updated Guide for 2026]

What Are AI Agents? [Updated Guide for 2026]

Table of contents

An AI agent can perceive context, reason about goals, make decisions, and take autonomous action without a human needing to micromanage each step. Team Gmelius explains how agents work in this article.

Written by

Anwesha Roy

Reviewed by

Florian Bersier

Last updated:

March 31, 2026

Expert Verified

Reading time:

14 min read

We've been tracking the agentic AI space closely, and the shift is unmistakable. Anthropic's Model Context Protocol (MCP), sometimes called the "USB-C for AI," has become a de facto standard for connecting AI agents to external databases, APIs, and enterprise tools. OpenAI and Google have embraced it. The Linux Foundation is now stewarding its development. And suddenly, the friction that kept AI agents trapped in demo mode is dissolving fast.

But if you're still asking "what actually is an AI agent?" you're not alone. The term has been thrown around so loosely that it has started to lose meaning.

So we went back to fundamentals, tested the latest platforms, and put together this updated guide to cut through the noise.

What Is an AI Agent? Definition

An AI agent is a software system that can perceive its environment, reason about goals, make decisions, and take autonomous action, without requiring a human to micromanage each step.

That last part is the key differentiator. A chatbot waits for your prompt and gives you text. An AI agent takes your high-level instruction and goes off to do the work.

Think of it this way: when you ask ChatGPT to draft an email, it generates text and hands it back. When you ask an AI agent to handle your inbox, it reads your messages, drafts context-aware replies trained on thousands of your previous conversations, routes emails to the right teammates, schedules follow-ups on your calendar, and flags only the stuff that truly needs your attention.

At the architectural level, most AI agents combine a foundation model (typically a large language model like GPT-4o, Claude, or Gemini) with memory, tool access, and a planning loop.

The agent perceives input, reasons through a chain of thought, selects the right tool or action, executes it, evaluates the result, and iterates until the task is done. It's this perception-reasoning-action loop that distinguishes a true AI agent from a static model sitting behind an API.

Inside AI Agents Working: Explained with an Example

Let's make this concrete. Say you're a sales manager, and a warm lead just replied to your outreach email at 11 p.m. You're asleep. Here's what an AI agent does, step by step:

  • Step 1: Perception - The AI agent detects the inbound email in your shared inbox. It reads the message, identifies the sender, and pulls context from your CRM: the lead's company size, the deal stage, and the last three interactions.

  • Step 2: Reasoning - Using its language model backbone, the AI agent determines that the lead is asking about pricing for an enterprise plan and wants to schedule a demo call. The agent cross-references your calendar, finds open slots, and considers your timezone and the lead's timezone.

  • Step 3: Action - The AI agent drafts a personalized reply with accurate pricing information pulled from your knowledge base, suggests three available time slots for a demo, attaches a relevant case study, and sends the email, all while you sleep.

  • Step 4: Evaluation - The agent logs the interaction in your CRM, creates a follow-up task for you to review in the morning, and notifies your team on Slack that the lead has moved to the next pipeline stage.

That entire workflow, which would normally eat 15 to 20 minutes of a human's morning, happened autonomously. No prompt required. No one was sitting at a keyboard. The AI agent perceived, planned, acted, and self-corrected. That's the agentic loop in practice.

There Are 3 Types of AI Agents Based on Use

Not all AI agents are built the same. We find it useful to break them into three categories based on how they interact with the world.

What Are AI Chat Agents?

AI chat agents are the most common type and the one most people encounter first. These are text-based AI agents that live inside messaging platforms, email clients, or dedicated interfaces. They communicate through written language — reading your messages, understanding context, and responding with text or by triggering backend actions.

Gmelius's Meli is a good example. This AI chat agent lives natively inside Gmail, drafting replies, sorting incoming email, and dispatching conversations to the right team members, all via text-based interactions.

‍

Install Gmelius (Pay Later)

Lindy AI works similarly: you describe a task in natural language, and the AI agent builds and runs the workflow for you.

What Are AI Voice Agents?

AI voice agents take the same perception-reasoning-action loop and apply it to spoken conversation. These AI agents can conduct phone calls, handle customer inquiries, schedule appointments, and even run sales outreach through natural voice dialogue.

We're seeing this category explode in 2026. DiligenceSquared, a Y Combinator startup covered recently, uses AI voice agents to conduct customer interviews for private equity due diligence, work that previously cost firms half a million dollars when outsourced to McKinsey or BCG.

‍ Lindy AI also offers voice agents through its Gaia platform, supporting over 30 languages for inbound and outbound calls.

What Are Multimodal AI Agents?

Multimodal AI agents can process and generate across text, images, audio, video, and even spatial data. They're the most ambitious category and the fastest-growing one.

Luma recently launched Luma Agents, built on its Unified Intelligence model family, which coordinates end-to-end creative workflows across text, image, video, and audio. These AI agents can plan a marketing campaign, generate ad creative, produce voiceover, and iterate on visual output, within a single agentic session.

OpenAI's Computer-Using Agent (CUA), which powers the agent mode now integrated into ChatGPT, is another example: it perceives screens visually, reasons about GUI elements, and takes actions by clicking, scrolling, and typing, just like a human at a computer.

AI Agents vs. LLMs (e.g., ChatGPT): How Are They Different?

This is the question we get asked most, and the confusion is understandable. ChatGPT is powered by a large language model, and ChatGPT now has agent capabilities. So where's the line?

Here's how we think about it. An LLM is a brain. An AI agent is a brain with hands, eyes, memory, and a to-do list.

A raw LLM takes a prompt as input and produces a completion as output. It's stateless, which means it doesn't remember previous interactions unless you feed them back in. It doesn't have access to your tools, and it can't take action in the real world. It's powerful, but passive.

An AI agent wraps that LLM in an architecture that gives it persistence (memory across sessions), perception (reading emails, seeing screens, hearing voice), tool access (APIs, databases, browsers, calendars), and autonomy (the ability to plan and execute multi-step tasks without human intervention at each step).

The practical difference matters enormously.

When we tested a standalone LLM against an AI agent built on the same underlying model, the agent completed a multi-step research and scheduling task in under four minutes. The LLM required seven separate prompts and manual copy-pasting between apps to accomplish the same thing. Same brain, wildly different outcomes.

Building Your Own AI Agent: Is It Worth It?

The honest answer: it depends on what you need and how much control you want.

Building a custom AI agent (like building an AI assistant) gives you full control over the reasoning logic, tool integrations, data flow, and security perimeter. If you're an enterprise with proprietary workflows, sensitive data, or edge-case requirements, building in-house makes sense.

The tooling has matured dramatically: you no longer need a PhD in machine learning to spin up an AI agent.

But the trade-offs are real. We've talked to teams that spent months building custom AI agents only to discover that an off-the-shelf tool handled 90% of their use case. Development time, maintenance burden, and the rapid pace of model improvements mean your custom agent can become outdated fast. Read about the cost of building AI assistants to understand this trade-off better.

Our recommendation: start with a pre-built AI agent that covers your primary workflow. Once you've identified the specific gaps, then consider building custom agents for those narrow, high-value use cases.

5 Best AI Agent Builders

If you've decided to build, here are the platforms we'd recommend evaluating in 2026.

1. Vertex AI (Google Cloud)

Google's Vertex AI has become a serious contender for enterprises that need to manage multiple AI agents in complex environments. It offers pre-built agent templates, integration with Google's ecosystem (BigQuery, Cloud Storage, Workspace), and robust MLOps tooling for monitoring agent performance in production.

Vertex AI (Google Cloud)

The managed infrastructure means you're not wrestling with server provisioning, and Google's latest Gemini models are available natively. If your organization already runs on Google Cloud, Vertex AI is the path of least resistance for deploying production-grade AI agents.

2. LangChain and Langflow

LangChain has become the open-source backbone of the AI agent ecosystem. It provides modular building blocks (chains, tools, memory modules, and retrieval pipelines) that let developers assemble custom AI agents from composable parts.

The LangSmith observability platform, which Replit notably uses for debugging its own agents, gives developers trace-level visibility into agent decisions and failures. You can also use it with Langflow, the AI agent builder that works on top of LangChain.

Langflow, which works on top of LangChain

LangChain is best for technical teams that want maximum flexibility and don't mind writing code. The community is massive and the ecosystem is rich.

3. Agent Builder by OpenAI API

OpenAI's API platform now supports native agent construction through the Assistants API and the Computer-Using Agent model. Developers can create AI agents with persistent threads (memory), code interpretation, file search, and function calling — plus the new CUA capability for browser-based task execution.

Agent Builder by OpenAI API

The Responses API has become the go-to for building agents that need to reason, act, and iterate. If you're already invested in the OpenAI ecosystem, this is the fastest route from prototype to production.

4. watsonx Orchestrate (IBM)

IBM's watsonx Orchestrate targets the enterprise segment with pre-built skill sets for HR, procurement, IT, and finance. The AI agent builder lets business users assemble workflows using natural language, while developers get API-level access for custom integrations.

​​

watsonx Orchestrate (IBM)

The platform's strength is governance: audit trails, role-based access, and compliance controls that matter in regulated industries. For companies that need their AI agents to meet strict enterprise requirements, watsonx Orchestrate remains a compelling choice.

5. Yellow Agent Builder (Yellow.ai)

Yellow.ai has carved out a niche in customer experience automation with its AI agent builder platform. It supports text and voice-based AI agents deployed across 35+ messaging channels and 135+ languages.

Yellow Agent Builder (Yellow.ai)

The no-code builder makes it accessible to CX teams without deep technical resources, while the underlying Dynamic AI Agents framework handles intent recognition, context management, and handoff to human agents when needed. For customer-facing AI agent deployments, particularly in multilingual environments, Yellow.ai is worth serious evaluation.

Bypass the Build: 5 Powerful AI Agents to Try for Free

Not everyone needs to build from scratch. These six tools deliver real AI agent functionality out of the box, and you can try all of them without reaching for your credit card.

1. Gmelius AI

Gmelius has evolved from an email collaboration tool into a full-fledged AI agent platform for Gmail.

Its flagship AI assistant, Meli, operates as a proactive executive assistant, drafting replies trained on thousands of your past conversations, sorting and classifying incoming email with precision, dispatching messages to the right teammates in shared inboxes, and even scheduling meetings through conversational interaction.

Gmelius's AI agents are powered by Google's latest Gemini models, with strict data privacy commitments (your data is never used to train models).

Gmelius's AI agent, Meli

The platform also includes AI Automation Architects that design custom workflow automations from natural language descriptions. If your work lives in Gmail, Gmelius offers one of the most polished AI agent experiences available today, with a free trial to get started.

Install Gmelius (Pay Later)

2. Reclaim.ai

Reclaim is an AI-powered scheduling agent that sits on top of your Google Calendar or Outlook and autonomously manages your time.

Reclaim t actively defends focus time, reschedules lower-priority events when conflicts arise, auto-schedules tasks based on deadlines and energy levels, and finds the best meeting times across multiple attendees' calendars.

Reclaim.ai

Acquired by Dropbox in 2024, Reclaim now serves over 320,000 users and reports saving an average of 7.6 hours per week per user.

The AI agent's priority-based scheduling intelligence (where it automatically bumps P4 events to accommodate P1 commitments) is genuinely clever. The free Lite plan is generous enough for individual use.

3. Saner.ai

Saner.ai positions itself as an AI productivity agent built specifically for knowledge workers drowning in context-switching — and especially those with ADHD.

It unifies notes, emails, tasks, and calendar events in a single interface, with a personal AI assistant called Skai that organizes your information, suggests tags, connects related ideas, and proactively builds a daily plan each morning based on your inbox, notes, and deadlines.

Saner.ai

The semantic search feature lets you ask questions in natural language and get answers pulled from your own data, even if you can't remember exact wording. The free plan includes 30 AI requests per month and 100 notes.

5. Operator by OpenAI (now ChatGPT Agent Mode)

OpenAI's Operator launched as a standalone browser-based AI agent in early 2025, then evolved into ChatGPT's integrated agent mode by mid-2025.

The concept is straightforward: you give the AI agent a task, and it opens its own browser to navigate websites, fill out forms, make purchases, book reservations, and handle repetitive web-based workflows on your behalf.

ChatGPT Agent Mode

Powered by the Computer-Using Agent (CUA) model, it perceives screens visually and interacts with GUIs by clicking, scrolling, and typing — just like a human would. It requests permission before sensitive actions and can hand control back to you at any time. The agent mode is now available to Plus, Pro, and Team users directly within ChatGPT.

5. Replit Agent

Replit Agent takes the AI agent concept and applies it to software development. You describe an application in plain language, and the AI agent plans the architecture, writes the code, sets up the database, debugs errors, and deploys the finished product, all inside Replit's cloud-based IDE.

Replit Agent

Agent 4, the latest version launched in early 2026, introduced a Design Canvas for visual planning, parallel task execution, and integrations with BigQuery, Linear, Slack, and Notion.

It supports 50+ programming languages and can build full-stack web and mobile applications from a single prompt. There's a free tier that includes your first 10 agent checkpoints, making it accessible for experimentation.

Understanding the Limitations of AI Agents

We would be doing you a disservice if we painted this as a frictionless utopia. AI agents in 2026 are powerful, but they come with real limitations that anyone deploying them needs to understand.

1. Hallucinations persist

AI agents inherit the hallucination tendencies of their underlying language models. An AI agent that confidently sends an email with fabricated pricing data or incorrect meeting details can damage client relationships fast. Human-in-the-loop review remains essential for high-stakes workflows.

2. Context windows aren't infinite

Even the best AI agents struggle with very long tasks that exceed the model's context window. Memory management (i.e., deciding what to remember and what to forget) is still an unsolved problem.

Trace, a Y Combinator-backed startup, is specifically attacking this "context engineering" challenge, arguing that whoever provides the best context at the right time will become the infrastructure layer for agentic AI.

3. Debugging is hard

When a traditional automation breaks, you can usually trace the failure to a specific step. When an AI agent fails, the reasoning chain is opaque. The agent might have misunderstood your instructions, hallucinated a fact, or made a bad judgment call in its planning loop. Observability tools like LangSmith help, but we're still early in developing robust debugging frameworks for AI agents.

4. Security is a real concern

Giving an AI agent access to your email, browser, and financial accounts creates an attack surface. Prompt injection, credential exposure, and unintended actions in production environments are active areas of security research for a reason. We strongly recommend starting with narrow permissions and expanding only as you build trust in the agent's behavior.

5. They're not always faster

For simple, one-shot tasks, opening a chatbot and typing a prompt can be faster than configuring an AI agent to handle it autonomously. Agents shine on repetitive, multi-step workflows where the setup cost is amortized over hundreds of executions. For a one-time task, sometimes the old-fashioned way is still quicker.

What Can You Do with Gmelius AI Agents?

Gmelius has built one of the most compelling AI agent experiences we've seen for email-heavy teams — and its flagship assistant, Meli, is the reason why.

Rather than bolting AI onto a separate app, Gmelius embeds Meli directly inside Gmail, where it functions as an autonomous personal secretary handling the coordination overhead drowning most professionals' days.

  • AI Meeting Scheduler: Instead of forcing contacts to an external booking page, the AI agent reads your emails, detects when a conversation needs a meeting, identifies the best slots based on your calendar patterns, and drafts a reply proposing times. One click to send. Rescheduling and cancellations? Handled.

  • AI Sorting and Dispatching: Meli tags every incoming email by content, context, and intent: action required gets flagged red, internal messages green, promotional noise auto-archived. For shared inboxes, the dispatching AI agent routes conversations to the right team member automatically. Unlike most competitors, the sorting categories are fully customizable.

  • Pre-Written AI Drafts: Meli identifies which conversations need replies, then generates drafts using your past email history, your learned tone of voice, and data from connected knowledge bases. The drafts wait in your inbox until you're ready to review and send. Custom prompts let you fine-tune the AI agent's behavior — tone preferences, per-domain rules, whatever fits your workflow.

  • Automatic AI Follow-Ups: The AI agent tracks unanswered emails that need replies (pending payments, stalled outreach, forgotten threads) and autonomously sends follow-ups after your chosen interval. No reminders to set. No tasks to forget.

  • Meli Chat: You can chat with Meli directly to summarize threads, fetch attachments, archive old emails, review your calendar, or even compose and send a new email, all without opening your inbox.

Gmelius hasn't gated Meli behind an enterprise paywall; the AI assistant for email is available from the entry-level plan, a sharp contrast to competitors like Fyxer and Superhuman that lock advanced AI behind tiers at $37+/month.

And with Gmelius headquartered in Switzerland, your calendar and communication data stay under some of the strongest privacy protections in the world.

Install Gmelius (Pay Later)

Are We Living in the Agentic AI Era?

We believe the answer is yes, but with caveats.

The infrastructure is finally in place. MCP gives AI agents a universal connector to the real world. Foundation models are powerful enough to reason through multi-step tasks. Browser automation lets AI agents interact with any website a human can. Voice capabilities are mature enough for real phone conversations.

But we're not in the "fully autonomous AI workforce" era that some breathless predictions promised. We're in something more interesting: the augmentation era.

The companies winning right now aren't the ones that hand everything to AI agents and walk away. They're the ones that carefully identify which workflows benefit from autonomy, maintain human oversight where it matters, and treat AI agents as force multipliers.

We're also seeing entirely new categories emerge. Agentic commerce — where AI agents browse, compare, negotiate, and purchase on behalf of consumers — is already prompting companies like World (Sam Altman's identity startup) to build verification systems that prove a human approves of an agent's purchasing decisions.

The bottom line: we've moved past the demo era. AI agents are in production, in inboxes, on phone calls, and inside codebases. The technology isn't perfect (we've outlined the limitations honestly) but it's functional, it's improving fast, and it's reshaping how work gets done.

If you haven't started experimenting with an AI agent yet, now is the time.

Try Gmelius for free.

‍

Share this post on LinkedinX facebook [#176]Created with Sketch.

Meet Meli, the AI Assistant who

Drafts your replies

Sorts your emails

Schedules your meetings

Dispatches emails to your teammates

\ Add Meli to Gmail](/content/ai-email-assistant/index.html)

Meet Meli, the AI Assistant who

Drafts your replies

Sorts your emails

Schedules your meetings

Dispatches emails to your teammates

\ Add Meli to Gmail](/content/ai-email-assistant/index.html)

Less Email. More Life.

Reclaim 1 hour per day managing email thanks to Gmelius shared inboxes and AI assistants built for Gmail.

\ Start your 7-day free trial](https://app.gmelius.com/signup) Get a Demo of Gmelius

More in

AI Assistants

How to Remove Hiver from Gmail](/content/blog/how-to-remove-hiver-from-gmail/index.html)

.avif)Superhuman AI Review: Is This Email Tool Worth It?](/content/blog/superhuman-ai-review/index.html)

.avif)The 11 Best AI Agents Available in 2026](/content/blog/best-ai-agents/index.html)

Fyxer AI Review: Can This AI Assistant Actually Manage Your Inbox?](/content/blog/fyxer-ai-review/index.html)

.avif)ChatGPT for Email Review: I Used ChatGPT to Write My Emails for a Week](/content/blog/chatgpt-for-email-review/index.html)

.avif)7 Best Google AI Writer Tools (All Compatible with Gmail)](/content/blog/7-best-google-ai-writer-tools/index.html)

Next

Product

Signup Demo Install Extension Install PWA for iOS & Android Install Google Workspace App API Documentation ⧉

Artificial Intelligence

Email Automation Agents

AI Draft Assistants

AI Sorting Assistants

AI Dispatching Assistants

Automation Rules & Agents

Team Collaboration

Gmail Shared Inbox

Shared Gmail Labels

Shared Email Drafts

Email Notes & @mentions

Email Delegation

Email Templates

Email Permalinks

Email Management

Kanban Boards

Email Analytics

Custom Views

Email Tags

Recent Conversations

Company

Pricing Legal Hub Partner Program Trust Center ⧉ Product Updates Product Roadmap ⧉

Customer Stories

French Red Cross

Grampian Lettings

The Harvard Crimson

Quimby Digital

Diane Endeavors

Garrison Forest School

Pod Forward

Norsemaison

Nori

Merserwis

Boise State University

Compare Gmelius with

Hiver

Fyxer AI

Superhuman

Front

Missive

DragApp

Keeping

SaneBox

Follow Gmelius

\ LinkedIn](https://www.linkedin.com/company/gmelius/) \ Instagram](https://www.instagram.com/gmelius/) \ X (Twitter)](https://x.com/gmelius) \ Youtube](https://www.youtube.com/channel/UC6HKJKctNXGihVjPOyNlWOg?sub_confirmation=1)

Resources

Blog Live Support Help Center ⧉ Gmelius Academy ⧉ Videos Status Page ⧉

Email Templates

Accounts Payable

Accounts Receivable

Admissions

Billing

Customer Service

Human Resources

Legal

Maintenance

Recruitment

Sales

Google Workspace Logos

Gmail

Calendar

Chat

Docs

Drive

Forms

Meet

Sheets

Slides

Integrations

\ Gmail](/content/integrations/gmail/index.html)

\ Google Groups](/content/integrations/google-groups/index.html)

\ Google Workspace Admin](/content/integrations/google-workspace/index.html)

\ Google Meet](/content/integrations/google-meet/index.html)

\ Google Calendar](/content/integrations/google-calendar/index.html)

\ Salesforce](/content/integrations/salesforce/index.html)

\ HubSpot](/content/integrations/hubspot/index.html)

\ Slack](/content/integrations/slack/index.html)

\ Trello](/content/integrations/trello/index.html)

\ Nicereply](/content/integrations/nicereply/index.html)

\ Zapier](/content/integrations/zapier/index.html)

\ Make](/content/integrations/make/index.html)

\ Relay.app](/content/integrations/relay-app/index.html)

By Gmelius

\ Backup Space ⧉](https://backupspace.io/?ref=gmelius)

Language

English

Français

Deutsch

Português

Español

日本語

Less Email. More Life.

© Gmelius. All rights reserved.

Terms of Service

Privacy Policy

Security

GDPR and DPA

Cookie Policy

AI Email Assistant for Gmail - YouTube

Tap to unmute

AI Email Assistant for Gmail Gmelius

Gmelius654 subscribers

Gmelius Info

I'm Gmail and I'm Meli for Gmail - YouTube

Tap to unmute

I'm Gmail and I'm Meli for Gmail Gmelius

Gmelius654 subscribers

Discover AI Assistants for Gmail

Welcome back, there! 👋

Here's what we added while you were away.

April Updates: Smarter Meli, lighter inbox 🌼

Hi there,

Say hello to more free time this Spring!

Get Meli to output exactly what you want

Gmelius gives users granular control over AI-generated replies. More professional? No problem. Mirror the sender’s tone? Easy peasy.

And now, custom instructions can contain up to 2,000 characters, up from 1,000.

This means more flexibility to fine-tune behaviour and outputs.

Meli now also looks at the last email in a thread to decide if it needs a reply.

So, if the last email includes the user in a CC/BCC or the last email is a promotion/notification, Meli can detect and ignore it. 🤫

Auto-schedule just the meetings you need, no noise

Meli understands meeting intent based on emails and generates auto-replies with available slots. To turn this feature on, go to Personal inbox configuration settings → AI Scheduler → Toggle to “Enabled”.

Now, the AI auto-scheduler turns itself off for promotional emails, messages where you are CC/BCC’d, and emails originating from Google Calendar.

Enjoy a cleaner calendar, with Meli by Gmelius.

What’s next? Catch up on a thread at a glance

In the next few weeks, we are launching the “conversation context summary” feature.

Users can click on the AI button at the top of any email for a quick summary of the conversation so far.

Stay tuned! And while you wait, check out the latest video from Gmelius, where we look at why 2.5 million users have deleted ChatGPT.

Why 2.5 Million People Just Deleted ChatGPT - YouTube

Tap to unmute

Why 2.5 Million People Just Deleted ChatGPT Gmelius

Gmelius654 subscribers

Keep freeing up time with Gmelius, and see you in May.

— The Gmelius Team

Meli is now your personal secretary. 📆 Meetings, handled.

New

March is all about reducing friction in your inbox and helping you move faster without extra effort. ✨

From smarter scheduling to improved organization and real-time sync, here’s what’s new in Gmelius, and what Meli can now do for you.

🤖 AI Meeting Scheduler by Meli: No More Back-and-Forth

When an incoming email contains a meeting request, Meli automatically detects it and acts as your personal secretary. Meli checks your calendar availability, applies your scheduling preferences, and drafts a ready-to-send reply with suggested time slots.

The recipient can then book a time in one click.

What makes it different: This is not chat-based scheduling. You don’t need to prompt it, Meli acts when it matters.

The AI Meeting scheduler is available for all customers. Learn more

PausePlay

-0:00

UnmuteMute

Disable captionsEnable captions

Settings

AudioEnglish

English

CaptionsDisabled

Disabled

QualityAuto (480p)

Auto

1080p

720p

480p

240p

SpeedNormal

0.25x

0.5x

0.75x

Normal

1.25x

1.5x

1.75x

2x

Exit PiPEnter PiPAirPlayExit FullscreenEnter Fullscreen

🏷️ AI Labels: Smarter Inbox Organization

Personal inboxes with AI Sorting enabled now generate Gmail labels instead of Gmelius tags, making it easier to organize and navigate conversations.

This update is now available across all plans.

What this means for you: Conversations are automatically categorized into clear Gmail folders creating an improved visibility and structure in your inbox both on desktop and mobile.

PausePlay

-0:00

UnmuteMute

Disable captionsEnable captions

Settings

Audio

CaptionsDisabled

Disabled

QualityAuto (480p)

Auto

1080p

720p

480p

240p

SpeedNormal

0.25x

0.5x

0.75x

Normal

1.25x

1.5x

1.75x

2x

Exit PiPEnter PiPAirPlayExit FullscreenEnter Fullscreen

⚡ Real-Time Updates Are Back

Your workspace now stays perfectly in sync without manual refresh.

We’ve improved real-time updates across key areas like:

  • Ticketing folders

  • Custom views

  • Shared workflows

Everything updates instantly, so you’re always working with the latest information.

🧹 Auto-Cleanup for AI Drafts

Meli now automatically deletes unused drafts to keep your Gmail drafts folder clean. AI-generated drafts are deleted after 3 business days if unused, helping keep your inbox clutter-free without any manual effort.

From the Editorial Team

On YouTube, we’re exploring whether AI truly understands your feelings… or is it just very good at pretending?

The AI Empathy Machine - YouTube

Tap to unmute

The AI Empathy Machine Gmelius

Gmelius654 subscribers

From smarter meeting scheduling to automatic cleanup, Meli keeps taking work off your plate.

And we’re only getting started. More updates coming soon.

- The Gmelius Team

February is our Favorite F-Word

New

This month might be 28 days short, but for Gmelius, it’s packed with updates, enhancements, and new releases.

Brand New Dashboard and Smarter Navigation

Your Gmelius dashboard is now arranged into four primary sections:

  1. Config for admins

  2. Mail for your inbox

  3. Meli to chat with your AI executive assistant

  4. and Analytics for monitoring productivity

Discover your new dashboard

(Even) More Control Over AI Drafts

Customizing AI-generated drafts is now easier with pre-built prompts. You can also make drafts more informed by adding multiple knowledge sources. Just add the websites from which you want Gmelius AI to fetch information, and you’re good to go.

AI Follow-Ups Are Here!

Finally, an AI that follows up for you. Meli (Gmelius AI) understands when a message you send needs a reply.

If the recipient doesn’t respond in 2 days (excluding weekends), Gmelius AI writes a follow-up for you to send in 1 click. If you get a reply, then no follow-up is generated. Simple!

You can configure AI follow ups directly from the AI Drafts tab.

FYI Tags to Identify Emails that Matter a Little Less

Tired of scrolling through long threads, only to find out you were just CC’d on it? Next time, check for the “FYI” tag in your inbox. It tells you when you’re CC’d or BCC’d into a conversation, so you can prioritize better and save time.

You can configure the new FYI tag directly from the AI Sorting tab.

Good News for 🇯🇵 Japan and 🇮🇹 Italy!

The Gmelius dashboard is now fully localized in Japanese and Italian. So, if you have friends who speak these languages, refer Gmelius and contact our team to earn referral credits :)

From the Editorial Team

On YouTube, we’re exploring how AI works as a massively influential force shaping our modern world.

The Coal Puzzle: Why AI Isn't Killing Jobs Yet - YouTube

Tap to unmute

The Coal Puzzle: Why AI Isn't Killing Jobs Yet Gmelius

Gmelius654 subscribers

We also created resources to help you improve productivity while saving time. Check them out:

📨 15 Best AI Assistants for Email Productivity: A Buyer’s Guide

✍️ These Are the 5 Best Email Generators of 2026

Product Updates: AI Sorting, Dispatching, Follow Ups & New Auto-Responder

January is about starting the year with a cleaner inbox and less manual work. Here’s a quick look at what’s new in Gmelius and what’s coming next, including multiple new skills added to Meli, your AI executive assistant.

✨ AI Sorting by Meli: Keep Non-Essential Emails Out of Your Inbox

For each AI category, you can now choose whether new emails should:

  • Stay visible in your inbox, or

  • Be automatically archived as soon as they’re categorized, and get closer to inbox zero thanks to Meli.

How it works in practice

If you set the Promotions category to Archive, emails identified as promotions won’t appear in your inbox. They’ll be automatically archived instead, keeping your inbox clear without losing access to the message.

Create your personal workspace or update its current configuration by going to the AI Sorting tab: https://app.gmelius.com/workspaces

This AI skill is only available for personal inboxes, not shared inboxes.

✨ AI Follow Ups by Meli: Get the "Ping" out of your Head

Take follow-ups off your mind and focus on the work only you can do. Meli stays on top of your inbox by tracking threads and crafting timely follow-ups.

AI Follow ups can be found and enabled in the AI Draft tab of your workspace under advanced settings: https://app.gmelius.com/workspaces

✨ AI Dispatching by Meli: Assign Conversations to the Right Teammate

Incoming conversations don’t always land with the right person. Meli changes that.

When accessing the new AI Dispatching tab, you’ll be able to define who should receive the conversation once it’s detected, and Meli will automatically route each message to the teammate best suited to handle it.

The result:

  • Less manual sorting

  • Faster response times

  • Better ownership across the team

This AI skill is only available for shared inboxes, not personal inboxes.

🪄 Set Up Autoresponders in Seconds

We’ve added a lightweight autoresponder rule to make simple automations easier.

On our Growth, Pro, and Enterprise plans, you can now quickly create automatic replies for out-of-office messages or instant acknowledgements, without building complex rules.

A faster way to stay responsive, even when you’re offline.

This is just the beginning of what we’re building this year…

Coming very soon:

  • A brand new look and better (user) experience for app.gmelius.com

  • Meli becomes your secretary by proactively scheduling meetings on your behalf.

I'm Gmail and I'm Meli for Gmail - YouTube

Tap to unmute

I'm Gmail and I'm Meli for Gmail Gmelius

Gmelius654 subscribers

.\. The Gmelius Team

Smarter Workspaces and new AI Controls

We're kicking off this month with some exciting news! We're premiering our first-ever micro-series, OOO (One Ordinary Office), on November 19. Get ready for a laugh—#AI just made its first workplace comedy. Check out the official trailer below!

One Ordinary Office | Official Trailer | Gmelius #gmelius #emailmanagement - YouTube

Tap to unmute

One Ordinary Office | Official Trailer | Gmelius #gmelius #emailmanagement Gmelius

Gmelius654 subscribers

Now, on to the product updates! We’re excited to roll out some powerful, much-needed updates designed to make your Gmelius experience even smoother. This month is all about giving you more control and smarter automation, starting with a brand-new command center for your team.

Meet Your New Command Center: The Gmelius Workspace

We've centralized your most important settings to make management a breeze. You can now access key information and settings from one convenient place. The Workspace is the new home for a host of powerful settings, including everything from your AI systems and integrations to your automation rules.

Here’s how to get started:

  1. Go to app.gmelius.com/workspaces

  2. Select your preferred Shared Inbox from the Gmelius homepage.

  3. Once inside, you'll find all the new Workspace settings on your left-hand side.

Please Note: Access to specific Workspace settings may depend on your role, such as owner or admin.

Automate Your Triage with Smarter AI Sorting

The Gmelius AI Sorting assistant is now even easier to use.

From your new Workspace, just click on AI Sorting in the left pane to see all available rules. Enable the rules you want to apply, and then use the magic wand button to automate how different types of emails are managed. Imagine being able to:

  • Auto-archive all promotional or outreach messages.

  • Get a Slack ping for urgent team and collaboration alerts.

  • Auto-assign or route emails containing financial info to a specific person.

Enforce SLAs like a Pro

SLA-based auto-assignment has been totally revamped and is now way easier to set up on Gmelius.

You can now automatically assign new conversations or email tickets to the right person based on your company's SLAs. This feature uses our round robin and load balancer rules, or you can use external integrations like HubSpot or Salesforce.

Additionally, new SLA warnings let you define specific response time targets. You'll be immediately alerted if a reply is delayed, helping you keep every conversation on track.

The AI Draft Assistant got even smart(er)

Gain full control over when and where your AI assistant steps in. You can now control which domains need replies and which ones don’t.

Gmelius now includes whitelists and blacklists for the AI Draft Assistant. This allows you to control precisely when responses are generated, helping you cut down noise and save time with Meli, Gmelius AI.

🦃 Get Ready for the Holiday Season

With the holidays just around the corner, check out these helpful resources from the Gmelius team:

Introducing Meli: Your AI executive assistant

New

Announcement

We’re thrilled to invite you to an early public preview of Meli, your new AI-powered executive assistant from Gmelius.

More than an email helper, Meli is your proactive executive assistant, built to anticipate needs, take initiative, and streamline your day.

With Meli, you can:

  • Summarize unread emails and flag what matters most.

  • Suggest archiving the noise and draft replies where action is needed.

  • Manage your calendar: events, reschedules, reminders.

  • Streamline daily email management so you can focus on real work.

With deep Gmail expertise and native Google Workspace integration, Meli is designed to help professionals reach inbox zero faster, cut response times, and focus on meaningful work.

💬 Start chatting with Meli

This is an early preview—and your feedback will directly shape how Meli evolves. Whether it’s helping you prepare for your day while commuting, managing incoming emails, or assisting your team, we’d love to hear how you’d use Meli:

Share feedback about Meli

Gmelius now speaks German, Spanish, and Portuguese! 🌍

New

Release: Gmelius now speaks German, Spanish, and Portuguese - YouTube

Tap to unmute

Release: Gmelius now speaks German, Spanish, and Portuguese Gmelius

Gmelius654 subscribers

¡Hola! Hallo! Olá!

We’re excited to announce that Gmelius just got more global. Our extension for Gmail and our Progressive Web App (PWA) are now available in German, Spanish, and Portuguese—joining our existing English and French versions.

Whether you're managing a shared inbox in Madrid, automating workflows in Munich, or collaborating with your team in São Paulo, Gmelius now works in your language.

🌐 Why it matters:

This update completes our first set of supported languages and brings Gmelius closer to teams around the world—offering a more intuitive and localized experience, wherever you work.

💬 How to switch languages:

Gmelius will automatically adapt to the language set in your Gmail or browser settings. No manual changes needed!

🎯 What’s next?

We’re not stopping here. More languages and localizations are coming soon to support our growing global community.

As always, thank you for using Gmelius to reinvent the way you work—wherever you are, and now, in your own language.

Powered by Featurebase