On April 21, 2026, Nous Research shipped Hermes Workspace Mobile, a phone app for their open multi-agent platform. It brings live tool runs, persistent memory, a skills catalog, a terminal, and recursive subagent orchestration to a device that fits in your pocket. This guide walks through what it is, why mobile matters for AI agents, and what it means for your automation stack.
Until recently, serious AI agent work meant sitting in front of a laptop. You opened a terminal, wired up tools, watched logs, approved actions. The new Hermes Workspace Mobile app flips that around. Long-running jobs keep moving while you are on a train, at a customer site, or on the couch. And the same agent that orchestrates subagents on your workstation is now a notification away on your phone.
The 30-Second Version
What Is Hermes Workspace Mobile?
Hermes Agent is Nous Research's open agent stack. It has crossed 50,000 stars on GitHub and grown into a local-first platform with memory, skills, permissions, and subagent coordination. The v0.9.0 release that landed this week adds two things that matter together: deeper subagent orchestration, and a first-class mobile app to watch and steer it.
The mobile app is not a trimmed-down chat view. It mirrors the desktop experience and adds phone-native touches like push notifications when a long job finishes, biometric approval for risky actions, and offline memory browsing.
Why Put a Multi-Agent Orchestrator on Your Phone?
Agent workflows are slow in wall-clock time. A research agent might crawl citations for an hour. An on-call agent might wait 40 minutes before a test suite finishes. A sales agent might grind through 200 prospects overnight. None of that work needs you in front of a screen, but all of it benefits from being reachable when the agent hits a permission prompt or finds something worth looking at.
Mobile solves three real problems that desktop did not:
- You can approve a tool call from a coffee shop without opening a laptop.
- You can glance at what a subagent is doing between meetings.
- You stop losing minutes to tab-switching and context rebuilds.
This matches the broader shift we wrote about in the orchestration era of agentic coding: the unit of work is no longer a single prompt, it is a long-running coordinator spinning up subagents, and you want to be able to check in from anywhere.
The Six Things the App Actually Does
Every feature in Hermes Workspace Mobile maps to something the desktop client already had. The point is that nothing important was left out.
1. Chat
A standard agent chat with streaming responses, history, and thread pinning. What matters is that the chat sits on top of the same session your desktop is in, so you can start a task at your desk, walk out, and keep the conversation going from your phone.
2. Live Tool Execution
The app shows every tool call in real time: which tool fired, what arguments it got, what it returned. Sensitive calls pause and ask for approval. Approving uses Face ID or Touch ID, so it takes a second and still keeps you in control.
3. Memory Browser
Hermes keeps a persistent memory graph across sessions. The mobile memory browser lets you search, pin, and delete entries. This is the part that turns the agent from "a chatbot that forgets" into "a coworker that remembers last Tuesday".
4. Skills Catalog
Skills are reusable recipes the agent has already learned: write a release note, triage an inbox, scrape a competitor's pricing. The mobile catalog lets you trigger a skill by name, pass a short input, and watch it run while you do something else.
5. Terminal
A read-mostly view of the agent's shell. You see what it is executing, and you can type commands if you need to nudge it. On a phone it is cramped, but for a quick tail -f or status check it is the fastest path.
6. File Inspector
Hermes runs in a sandboxed workspace with its own filesystem. The inspector lets you open diffs, download artifacts, and delete cruft from the phone. If your agent wrote a report, you can read it right there instead of SSHing in.
The Real Story: Wider and Deeper Subagent Orchestration
The mobile app is the headline, but the orchestration upgrade under it is what changes how the system is used. Hermes subagents now support both greater spawn width and recursive spawn depth. In plain terms:
- Width: the top-level agent can fan out many parallel subagents for independent workstreams.
- Depth: each subagent can spawn its own subagents, and so on, several levels deep.
That is how you get real hierarchical decomposition instead of one big agent doing everything serially. A research task can split into ten parallel readers, each of which spawns a summarizer and a fact-checker. A coding task can split per file, per module, or per failing test.
Agent frameworks are quietly converging on the same shape: a durable coordinator, a memory layer, a skills registry, and recursive subagents. Hermes is now one of the cleanest open implementations of that pattern.
Hermes Workspace Mobile vs ChatGPT Mobile vs Codex CLI
These three are often compared, but they solve different problems. Here is the honest breakdown:
| Feature | Hermes Workspace Mobile | ChatGPT Mobile | Codex CLI |
|---|---|---|---|
| Open source | Yes | No | Partial |
| Runs locally | Yes | No | Yes |
| Persistent cross-session memory | Yes (graph) | Limited | No |
| Recursive subagents | Yes (width + depth) | Limited | No |
| Live tool execution view | Yes | No | Terminal only |
| Skills catalog | Yes | Custom GPTs | No |
| Mobile push approvals | Yes (biometric) | No | No |
| Best for | Multi-agent automation on the go | Chat and quick tasks | Local coding sessions |
Not a Replacement for Your Laptop
Five Practical Use Cases
The On-Call Engineer Agent
The agent watches alerts, opens a draft pull request with a proposed fix, and pings your phone. You approve or reject from the app. You never open your laptop unless the fix actually needs your head in it.
The Sales Researcher
Before every meeting, a subagent pulls news, funding, and team changes for the company you are about to see. It sends a one-page brief to the app as a push notification, so you read it on the way in.
The Data Pipeline Monitor
A parent agent orchestrates subagents that check cron jobs, queue depths, and cost budgets across environments. If anything crosses a threshold, it shows up as a single summary on your phone, not ten separate pages.
The Content Drafter
You paste a rough idea into the app. A skill kicks off, spawns a research subagent, drafts a 700-word post, and sends it back for your review. You edit on the phone if it is short, or send it to your desk if it is not.
The Personal Assistant
A small swarm of subagents handles calendar conflicts, inbox triage, and travel bookings. The mobile app becomes the single approval inbox for the whole stack, which is how a personal assistant is supposed to feel.
How to Get Started
- 1Install Hermes Agent on your workstation. The repository is on GitHub under the NousResearch organization.
- 2Pair the mobile app over the local network or a signed tunnel. Everything stays in your environment.
- 3Enable the tools and skills your use case needs, and set approval rules for anything that costs money or touches production.
- 4Start with one workflow. Get comfortable with push approvals before turning on wider subagent trees.
If you are completely new to this space, our OpenClaw 101 guide for new users walks through the concepts that also apply here: skills, permissions, and why an agent is not the same as a chatbot. If you are coming from the OpenAI side, compare notes with our post on ChatGPT Workspace Agents explained simply.
The Bottom Line
The AI agent story for 2026 is not bigger models. It is better harnesses: durable coordinators, persistent memory, clean subagent trees, permissioned tool use, and now, a mobile remote control. Hermes Workspace Mobile is a credible open answer to all of those at once, and it is the first time the orchestration story feels complete on a phone.
The interesting question is not whether to install it. It is which of your slowest workflows becomes bearable once an agent can run it for you while you walk the dog, and ping you when it needs a yes or a no.