Hermes Agent v0.17 surges: AI assistant grows new tentacles, iMessage, Raft, and back-end sub-Agents are all here

Hermes Agent v0.17.0 This update has been officially given a very accurate name: The Reach Release.

v0.16 puts Hermes on the desktop, and v0.17 solves another problem: How far can this AI assistant reach?

The answer is: reaching into iMessage, WhatsApp, Telegram, Raft Agent network; reaching into desktop apps, backend Subagent, picture editing, automation blueprints, skills market, memory system; and even reaching into the Composer programming model behind Cursor.

Starting from this version, Hermes is no longer just an open source Agent that can chat, remember, and perform tasks, but is heading towards a "personal AI operating system with multiple entrances, multiple tools, and multiple scenarios".

Let’s look at the version size first

The official release data is very exaggerated:

• Published: June 19, 2026
• Since v0.16.0: ~1475 commits
• Merged PR: ~800
• Changed files: 1693
• New code: 235,390 lines
• Deleted code: 50,730 lines
• Close issue: 300+
• Community Contributors: 245

This is not a minor fix, this is a truly major release. The official name is Reach Release, which is very appropriate: Hermes’ tentacles have begun to expand on a large scale.

iMessage access: Blue Bubble finally no longer needs Mac relay

One of the most exciting features this time is that Hermes officially connects to iMessage.

In the past, if you wanted the AI ​​assistant to send and receive iMessages, you basically couldn't get around a Mac: either put a Mac mini in the corner to run BlueBubbles, or you could use the bridging service yourself. It works, but the maintenance cost is annoying.

v0.17 directly supports the Photon Spectrum plug-in:

• No Mac relay required
• No BlueBubbles bridge required
• No need to host the relay yourself
• Log in to Photon to manage the line pool through device code

The command is also very straightforward:

hermes photon login

Once logged in, Hermes can send and receive messages in iMessage. For users whose family and friends live in a blue bubble, this means that Hermes can finally naturally enter their most commonly used chat portal.

Officials even position it as the successor to BlueBubbles: free to get started, no self-hosting required.

Raft Agent Network: Hermes begins to connect to other people’s agent world

The second very strategic update is the Raft platform adapter.

Raft is an Agent network, and Hermes v0.17 can be accessed as an external Agent through the wake-channel bridge. Set RAFT_PROFILE, run bridge, and Raft will wake up Hermes to process messages.

What's more critical is its privacy design: the wake payload only carries metadata, such as event ID and timestamp, without the message body.

This shows that Hermes not only wants to be a personal assistant, but also wants to become a node in the Agent network. In the future, your Hermes may not be an island, but a member that can be orchestrated, awakened, and coordinated by other Agents.

There's more to this than meets the eye. Because the real value in the Agent era is not how smart a single AI is, but how multiple Agents can collaborate safely.

Desktop App finally looks like a serious workhorse tool

When v0.16 first launched the desktop app, it was more like “finally there”. v0.17 is "finally able to be the daily main force."

This desktop enhancement includes:

• Rebindable shortcut key panel
• Native system notifications and support switching by type
• Sub-Agent real-time observation window can stream delegated tasks to an independent pane
• Composer model selector, supports different model presets
• Automatic RTL/bidi text orientation
• Resizable VS Code themed terminal panel
• Independent composer drafts per thread
• Install the VS Code Marketplace theme directly

In the past, the desktop was "one more entrance besides the terminal", but now it is more like a real Agent workbench. Especially live subagent watch-windows: you throw a long task to the sub-agent, and it works in real time in another window while your main line continues to advance.

This is not a UI chore, but a reconstruction of the interface for human and multi-Agent collaboration.

Backstage Subagent: Finally no more waiting around

v0.17 added background / async subagents.

In the past, if you entrusted a sub-Agent to do long tasks, such as researching a project, running a round of builds, and writing a report, the main conversation would easily get stuck. Now delegate_task(background=true) will immediately return a handle, the sub-Agent runs in the background, and you and the main model continue to do other things.

When the background task is completed, its complete results will automatically return to the main context as a new round of dialogue.

This may sound like a small feature, but it's a game-changer for complex tasks.

For example, you can arrange it like this:

• A sub-Agent searches information
• A sub-Agent runs the test
• A sub-Agent generates a comparison table
• The main agent continues to discuss the plan with you

This is true parallel collaboration of Agents, rather than an AI slowly queuing up to do things.

image_generate supports image editing

Hermes' image_generate tool was previously primarily responsible for generating images from text. Starting from v0.17, it can receive existing images, and then edit and convert them according to the prompt words.

Typical usage:

• Change this logo to blue
• remove background
• Turn sketches into product renderings
• Adjust the style according to the reference picture

It does not open a new image_edit tool, but follows the same pattern of image_generate and automatically routes to the edit endpoint of the backend provider. For Agents, this is important: the more unified the tool interface is, the simpler task orchestration will be.

Automation Blueprints: Automation can be done without cron

Hermes originally supports scheduled tasks, but cron is too unfriendly to ordinary users. The Automation Blueprints of v0.17 are designed to solve this problem.

You don’t need to memorize 0 8 * * *. Just select an automation blueprint, such as "News briefing at 8 o'clock every morning", and Hermes will directly ask you for the parameters you need.

The same blueprint can be rendered natively on multiple surfaces:

• Dashboard is a form
• In CLI/TUI are commands
• Conversations in Messenger
• The document is a directory entry

This shows that Hermes is doing a very smart thing: turning automation from "engineer configuration" to "ordinary people asking questions."

Cursor Composer models are also available

v0.17 also has an update that is very popular for programmers: grok-composer-2.5-fast is added to the xAI OAuth model selector.

Composer is the high-speed programming model behind Cursor. Now if you have an xAI Grok subscription, you can call it directly through Hermes, with the Context window aligned to the full 200k.

This is equivalent to stringing three things together:

• Grok Subscribe
• Hermes Agent loop
• Encoding speed of Cursor Composer

No separate API Key required. For developers who have already purchased Grok, this feature is equivalent to getting a high-speed programming backend for free.

Dashboard becomes a complete configuration center

The Dashboard of v0.17 is not just about viewing the status, but can also create a complete Hermes Profile.

You can directly in the browser:

• Select model
• Select Skills
• Mount MCP Servers
• Manage multiple profiles
• Global switch profile

In the past, these things required manual modification of config.yaml. No need now, it can be done with just a few clicks in the Dashboard.

This is critical for team deployment. Because once Agent is used by non-technical people, the configuration must be visualized.

Skills Hub reworked: skills installation finally looks like an app store

The Skills Hub has also been significantly changed this time.

New capabilities include:

• Connected hubs
• Featured Recommended Area
• Full preview of skill before installation
• Automatic security scans for each skill
• Supports trusted taps: OpenAI, Anthropic, HuggingFace, NVIDIA

This means that Skills is no longer a flat list, but begins to move towards a real Agent app store.

Big upgrade of memory tools: atomic batch operations

Hermes’ long-term memory has always been a core selling point. In v0.17, the memory tool adds an operations array, which can add/replace/remove in batches at once and execute it as an atomic operation.

In the past, if a model wanted to update its memory, it might take several steps: first delete old content, then add new content, and then deal with budget overflow. If any step in the middle fails, the memory may become confused.

Now it can be done in one call:

• free up space
• Replace old information
• Add new memories
• Also check the final character budget

This makes memory updating faster and less likely to fail halfway. For an Agent that “gets to know you better the more you use it”, this is a bottom-level stability upgrade.

Dashboard login reinforcement

In terms of security, v0.17 has strengthened Dashboard authentication:

• token-required endpoint correctly returns 401
• websocket uses served dashboard token authentication
• Warn when public_url override is silently rejected

This kind of change isn't cool, but it's very necessary. Once the Agent Dashboard is exposed to the network, it may be connected to models, tools, MCPs, file systems, and message channels. If there are holes in the authentication, the consequences will be much more serious than ordinary backends.

WhatsApp and Telegram are also more complete

In addition to iMessage, v0.17 also enhances WhatsApp and Telegram.

For WhatsApp, Hermes has added an official WhatsApp Business Cloud API adapter. In the past, you could use Baileys bridge, but the bridge process required maintenance, code scanning, and protection against disconnection. Now you can directly connect to Meta’s official hosting channel.

In terms of Telegram, Hermes supports rich messages of Bot API 10.1. The long message format is cleaner, and the native markup is more complete. It is enabled by default and can also be opt-out.

This shows that Hermes’ message entry is changing from “acceptable” to “received decently”.

Curator no longer secretly burns Tokens

There is also a very pragmatic optimization: Skill curator no longer calls aux-model in routine run to perform expensive consolidation by default.

Now only free deterministic inactivity sweep is done by default. Only if you explicitly turn on curator.consolidate: true, or run hermes curator run –consolidate, will tokens be spent for smarter skill merging.

In a word: daily backend organization skills, no longer secretly burning money.

Why v0.17 is said to be “tentacle expansion”

The core change in v0.17 is not a certain function, but the form of Hermes.

In the past, Hermes was like an AI assistant installed in the terminal.

v0.16 gives it a desktop.

v0.17 makes it start to extend its tentacles:

• Reach into the iMessage blue bubble
• Reaching out to the Raft Agent network
• Reaching out to WhatsApp official API
• Reach into Telegram rich text messages
• Reach into the sub-Agent window of the desktop app
• Dig into the photo editing and video/image generation process
• Dig into automation blueprints and codeless configuration
• Reach out to Skills Hub and Profile Builder
• Reach into the underlying update logic of the memory system

This is the true meaning of Reach Release: Hermes no longer waits for you to open the terminal to find it, but actively appears at various entrances to your work, chat, automation, and collaboration.

Who is suitable for upgrading?

If you are using Hermes Desktop, you must upgrade. The maturity of this desktop version has been significantly improved.

If you use Subagent heavily, you must upgrade. The background sub-Agent can directly change the experience of long tasks to a new level.

If you want to connect to iMessage, WhatsApp, and Telegram, you must upgrade. Message entrance enhancement is the main line of this version.

If you are deploying Hermes in a team, it is also recommended to upgrade. Profile Builder, Dashboard login reinforcement, and Skills Hub security scanning are all making up for the shortcomings of team use.

Summary

Hermes Agent v0.17.0 is not a version with “more functions”, but a version with “further boundaries”.

From iMessage to Raft, from desktop sub-Agents to automated blueprints, from picture editing to memory atomic operations, Hermes is upgrading itself from an AI assistant to a cross-entry, cross-tool, and cross-scenario Agent operation layer.

If v0.16 gives Hermes a body, then v0.17 gives it tentacles.

The AI ​​assistant no longer just sits in the chat box waiting for you to send messages, it begins to actively enter your messages, desktop, tool chain and automated processes.

This is what Agent should really look like.