Hermes Agent v0.17 brings async sub-agents, iMessage, and smarter automation

Hermes Agent v0.17 is a big step forward for people who actually want AI agents to do real work, not just chat. This release focuses on better automation, collaboration between agents, and easier control from your desktop and phone.

How to update to Hermes v0.17

You can update Hermes directly from the dashboard. Open Hermes, go to Manage, scroll down, and click Update Hermes. Once it finishes, your version number should show 0.17 instead of 0.16.

If you prefer the terminal, you can run hermes update to pull the latest version.

Hermes now works with iMessage

One of the headline features is native iMessage support. After you log in and authenticate, Hermes can send and receive iMessages on your behalf.

In practice, this means you can:

• Use your Hermes agent from your phone via iMessage
• Have Hermes respond to messages, follow up, or handle simple conversations
• Treat your agent more like a real personal assistant that lives inside your messaging app

Previously, getting an AI into iMessage was clunky. v0.17 makes it a first-class channel.

Raft integration: Hermes as a gateway agent

Hermes v0.17 also adds support for Raft, a platform where humans and AI agents can build together. Hermes can now join the Raft network as a gateway channel using a bundled Raft platform adapter.

What this means in simple terms:

• You can connect Hermes to Raft as an external agent
• Hermes can act as a bridge (or “wake channel”) between Raft and other tools
• It becomes easier to plug Hermes into multi-agent ecosystems and collaborative workflows

Improved Hermes desktop app

The desktop app has had a noticeable upgrade. Once you open Hermes Desktop, you’ll see an Update now button if a new version is available.

Key improvements include:

• Native OS notifications so you don’t miss agent updates or task completions
• Live sub-agent watch windows to see what your sub-agents are doing in real time
• General UI and usability improvements for managing agents and tasks

Async sub-agents: true parallel work

One of the most powerful changes in v0.17 is asynchronous sub-agents. Hermes can now spawn a team of sub-agents that work in parallel instead of one after another.

Here’s how it works conceptually:

• You have a lead agent (your main Hermes profile)
• You give it a complex task that can be split into parts
• Hermes automatically spawns multiple sub-agents, each handling a different part
• When they’re done, Hermes collects and merges their results for you

Example: you want a full content package for a new page on your website. You could ask Hermes to:

• Research a topic and keyword
• Optimize the page for SEO
• Publish the content to your site
• Create a supporting video

Hermes can spin up sub-agents for keyword research, article writing, video scriptwriting, and video editing at the same time. You see the final combined result in your chat instead of having to manage each step manually.

If you’re already using agent swarms, this async approach builds on ideas from earlier releases like Hermes Agent v0.15, but with more control and better performance.

New image editing skill (image-to-image)

Hermes now supports image editing, not just image generation. The new skill lets you transform a source image instead of always starting from scratch.

Capabilities include:

• Text-to-image generation
• Image-to-image editing and transformation

This works with multiple backends such as OpenAI, Grok, Krea, OpenAI Codex or OAuth-based connections, depending on your setup. In practice, you can ask Hermes to adjust, restyle, or extend an existing image as part of a workflow instead of using a separate tool.

Automation blueprints: scheduled workflows without coding

Automation blueprints are pre-defined task templates you can drop into Hermes to create scheduled automations without writing your own syntax.

Here’s how they work:

• You grab a blueprint from the official Hermes portal (for example, "News Research")
• You paste it into Hermes
• Hermes guides you through setup and scheduling

A typical blueprint is broken into:

• The task – what the automation is supposed to do
• The steps – the sequence Hermes will follow
• The output structure – how the final result should be formatted
• Style and constraints – e.g. under 600 words, 1–2 sentences per item

Example: an AI news digest blueprint might run every 7 days, gather the latest AI news, summarize each item in 1–2 sentences, and package everything into a short report. You don’t need to understand the underlying code; you just plug the blueprint into Hermes and customize a few details.

There are also more technical blueprints for things like fixing pull requests or running coding tasks, but non-technical users can still get value from simple research and reporting automations.

New models: Grok Composer 2.5 fast via X

If you have an X (Twitter) subscription, you can now use Grok Composer 2.5 fast inside Hermes via Cursor’s Composer model.

Setup is straightforward:

• In your terminal, run hermes model
• Scroll to the XAI Grok section
• Use Grok OAuth to log in
• Select Grok Composer 2.5 fast from the model list

This makes it easier to mix and match different LLMs inside Hermes, something that’s especially useful if you’re already exploring multi-model setups or advanced orchestrations (see also top AI models for Hermes Agent).

Profile builder: specialized AI personas for every task

Hermes v0.17 introduces a full profile builder inside the dashboard. Profiles are separate AI agent personas, each with their own skills, instructions, and even model choices.

What you can do with profiles:

• Create many different agent profiles (dozens if you like)
• Assign different APIs or models to each profile
• Give each profile its own skills and custom instructions
• Switch profiles in chat with a couple of clicks

Example setup:

• A Content Judge profile that reviews everything other agents write and scores quality
• A Video Director that plans video concepts and structure
• A Content Editor that polishes drafts and enforces your style guide
• Separate profiles bound to specific models like GLM 5.2, Jarvis, or Chimera K2.7

In a real workflow, you might have one profile generate a blog post and video script, another profile judge the quality (e.g. 4/10, 7/10, etc.), and a third profile revise until the judge is satisfied. The new profile builder makes this kind of multi-agent team much easier to set up and manage.

Better skill browser and management

The skill browser has been completely overhauled. You can now:

• Browse all available skills in one place
• Search by keyword
• Filter by category
• See which skills are installed and toggle them on or off

You can access skills via the terminal (for example, using hermes-order) or through an agent operating system UI if you’re using one. The new browser makes it much easier to understand what your agents can actually do and to keep your setup clean.

Memory and UI improvements

The memory system in Hermes has also been upgraded, giving you more control over how and what the agent remembers. This is especially useful if you’re running long-term projects or want Hermes to stay aware of your preferences across sessions.

Other smaller but useful updates include:

• A smoother dashboard login experience
• Official WhatsApp integration
• Rich text support for Telegram
• Token optimization and new UI options

Loop engineering: built-in quality control loops

One of the most interesting ideas in v0.17 is loop engineering. This lets you create automatic feedback loops between agents so quality control is handled by AI, not by you manually checking every output.

The basic pattern looks like this:

• You define a goal for Hermes (e.g. write a high-quality article, generate a plan, etc.)
• You set a starting point and choose a builder agent
• You set a judge agent and the number of rounds the loop can run
• The builder creates something, the judge scores it, and if it’s not good enough, it loops

The system works like this:

1. You set the quality bar (for example, a minimum score)
2. The builder agent produces work
3. The judge agent grades it and gives feedback
4. If the score is too low, the builder tries again with that feedback
5. The loop continues until the work passes the threshold

In a real run, you might see scores like 54 → 71 → 83 → 92, and only when it hits 92 does the loop stop and deliver the final output. This saves you from constantly reviewing drafts and giving feedback yourself—the agents handle the iteration.

Why v0.17 matters for serious AI workflows

Hermes Agent v0.17 is less about flashy demos and more about making AI agents practical for real work. With async sub-agents, robust profiles, automation blueprints, iMessage and Raft integration, and loop-based quality control, you can start to treat Hermes as an actual team of assistants rather than a single chatbot.

If you’re building complex workflows, running multiple models, or moving toward an “AI agent manager” role, this release gives you many of the building blocks you need to orchestrate agents at scale.