Hermes Agent is not just getting more features. It is becoming easier to reach, easier to inspect, and easier to shape into separate working agents. That is the useful story behind this update.
The walkthrough below comes from Alex Finn, who walks through Hermes Agent's newest update: native iMessage, background agents, Unreal MCP, a refreshed desktop app, profile builder, Skills Hub, smarter self-improvement, and richer Telegram messages. Credit to Alex for the walkthrough and practical framing. JQ AI SYSTEMS is adding the builder/operator interpretation: what should you actually do with it?
The update matters because Hermes is moving toward a reachable agent layer: desktop for control, iMessage for quick commands, Telegram for richer mobile output, profiles for separation, and background agents for parallel work.
Source note
The official Hermes Agent release notes list v0.17.0 as the "Reach Release." The notes align closely with Alex Finn's video: native iMessage through Photon, automatic background subagents, Unreal Engine 5.8 MCP, desktop improvements, a visual profile builder, Skills Hub, better memory edits, and Telegram rich formatting.
I am treating the release notes and official Hermes documentation as the factual spine. The video is valuable because it shows how a power user thinks about the update in practice.
What changed
The simplest way to read this update is that Hermes is expanding in two directions at once:
- More reach: iMessage and Telegram make Hermes accessible from the places people already send messages.
- More control: the desktop app, profile builder, Skills Hub, memory edits, and subagent views make Hermes easier to operate and inspect.
That combination matters. Messaging without control creates chaos. Control without reach stays trapped in the power-user corner. Hermes is trying to connect both.
Alex Finn's walkthrough covers
- Native iMessage: text Hermes from Apple's Messages app through Photon.
- Background agents: Hermes can automatically spin up subagents for complex work.
- Unreal MCP: agent workflows can connect into Unreal Engine tasks.
- Desktop app: better windows, model selection, live subagent tree, and terminal access.
- Profile builder: visual creation of separate Hermes profiles.
- Skills Hub: a browsable place to discover and install skills.
- Smarter self-improvement: Hermes updates its own memory and skills more often.
- Telegram updates: smoother streaming and richer formatted outputs.
iMessage and reach
Native iMessage support is the headline because it changes how approachable Hermes feels. Instead of opening a terminal, desktop app, or Telegram thread, you can message the agent from the default phone surface many people already use all day.
The useful mental model is not "iMessage replaces Hermes Desktop." It is more like this:
Quick reminders, simple briefs, light research requests, asking Hermes to queue a task, or checking a status while away from the machine.
Reviewing tools, inspecting subagents, editing profiles, watching terminal output, approving actions, and understanding what the agent actually did.
There is also a safety angle. Messaging channels are frictionless, which is exactly why they need boundaries. If Hermes can act on your behalf, your iMessage prompts should not become a back door to sending, deleting, spending, publishing, or changing production data without review.
Background agents
The official release notes describe automatic background subagents as a headline feature. In the video, Alex emphasizes that Hermes can now keep working in the background when a task is complex enough, while you continue using the main thread.
That is a real workflow upgrade. A single agent doing everything in one thread can become slow and crowded. Background agents make sense when the work can be split:
- research five competitors separately, then summarize the differences;
- inspect multiple files or modules in parallel;
- generate options for copy, code, design, or product positioning;
- test separate hypotheses before the parent agent recommends a path;
- run focused review passes for security, UX, performance, and documentation.
The risk is that parallel work can multiply bad assumptions. The parent agent still needs a merge step, evidence, logs, and human review.
Background agents are powerful when each worker has a narrow job and the parent agent has clear verification criteria. Without that, you get more output, not necessarily better output.
Desktop control room
Hermes Desktop is becoming the operator surface. The official desktop documentation frames it as a native app that shares the same Hermes core, config, API keys, sessions, skills, and memory as the CLI and gateway.
Alex's walkthrough highlights a few practical improvements:
- Separate windows: keep multiple work threads visible without losing context.
- Bottom model selector: switch model choice from the chat surface.
- Live subagent tree: see which child agents are running and what they are doing.
- Built-in terminal: inspect command output without leaving the agent workspace.
This matters because agent work needs observability. A better agent is not only one that takes more actions. It is one whose actions are visible enough that you can trust, debug, and improve the workflow.
Profiles and Skills Hub
Profiles are one of the most important Hermes concepts. Officially, profiles let you create separate agents with their own config, keys, memory, sessions, skills, cron jobs, and state. That is how one Hermes install can behave like several purpose-built assistants.
The new profile builder makes this less manual. Instead of forcing every user into config files, Hermes can help create a profile through the dashboard. That is the right direction. Agent architecture should be visible enough for non-terminal operators to shape.
The Skills Hub is the other big piece. The official skills documentation describes skills as reusable capabilities and procedures. A hub makes skills discoverable, but it also raises the review bar.
How I would use the Skills Hub
- Browse existing skills for ideas.
- Inspect what the skill can access and what instructions it adds.
- Install only the skills needed for one profile.
- Ask Hermes to create a smaller custom skill when a public one is too broad.
- Keep a review habit for skills that touch files, browsers, credentials, or external services.
Skills are not decorations. They are operational context. Too many broad skills can make a profile noisy, expensive, and harder to audit.
Self-improvement
The most agent-native part of the update is smarter memory and skill editing. In the video, Alex says Hermes now creates and updates its own skills more often, especially after repeated use. He describes Hermes patching and improving an Unreal MCP skill as it learns from attempts.
This is promising, but it should not be treated as magic. Self-improvement is only useful if the agent improves the right thing. A bad lesson preserved in memory can become a recurring bug.
For business use, I would add three rules:
- Review generated skills before trusting them. A self-written skill can encode a wrong assumption.
- Keep change notes. If a workflow suddenly changes behavior, you need to know what memory or skill was updated.
- Separate experiments from production profiles. Let Hermes learn in a sandbox before copying stable patterns into your main profile.
The best agents will learn. The safest teams will still review what they learn.
Telegram updates
Telegram remains important because it is better suited than iMessage for structured, threaded, and formatted interaction. The update improves streaming and rich formatting, including tables, lists, bold text, and more readable data responses.
That may sound small, but output format changes behavior. If an agent can return a clean table of stock research, project status, market notes, or lead triage in a mobile chat, people are more likely to use it as an operating assistant instead of a novelty bot.
My practical split would be:
- iMessage: fast personal commands and lightweight check-ins.
- Telegram: structured mobile work, tables, longer responses, and richer agent updates.
- Desktop: setup, review, debugging, profile work, terminal output, and high-risk actions.
Builder checklist
If you are trying this Hermes update, do not connect everything on day one. Start with one controlled workflow.
- Pick one profile. Give it a job, a model, a working directory, and a small skill set.
- Pick one mobile channel. Use iMessage for quick commands or Telegram for richer outputs.
- Pick one background-agent task. Use something reviewable: research, summary, test plan, or comparison.
- Define what done means. Tell Hermes what evidence to return, what files to change, and what to verify.
- Keep high-risk actions gated. No sending, spending, deleting, publishing, or production changes without approval.
- Review new skills and memories. If Hermes self-improves, inspect the changes before relying on them.
- Measure cost per useful output. More agents and more channels only matter if they produce better completed work.
The point is not to have the flashiest agent setup. The point is to turn repeated work into a controlled system that can be reached from the right place and reviewed from the right surface.
Do not turn on every new Hermes surface at once. Pick one workflow, give it a profile, one skill set, one review habit, and one messaging channel.
Sources
- Alex Finn walkthrough: Hermes agent's new update is amazing
- Alex Finn on X
- Hermes Agent official releases
- NousResearch/hermes-agent on GitHub
- Hermes Desktop docs
- Hermes Profiles docs
- Hermes Skills docs
- Hermes Subagent Delegation docs
The short version: Hermes is getting more reachable, but the builder advantage is not the messaging channel. It is the controlled workflow behind it: profiles, skills, background agents, logs, and review.