The Hermes Agent inbox earns a 7 out of 10 from me: a safe, zero-setup status view of your agent's unfinished business, held back by missing approvals and by having no maintainer.
It is the inbox labs plugin from Nous Research, version 0.1.0, and it has nothing to do with email.
The ratings below are my editorial judgement from the plugin README, the catalog entry and the Hermes docs, not from benchmark runs.
If you want my team to review your own agent setup, book a free SEO strategy session.
The Verdict In One Line
Install it if Hermes runs cron jobs, delegations or background work for you, and never treat an empty screen as proof that nothing is waiting on you.
The Hermes Agent Inbox Scorecard
| Criterion | Score | Reason from the docs |
|---|---|---|
| Setup | 9/10 | One install command, one new CLI command, and the configuration section says "None." |
| Safety | 9/10 | Files are read as files and SQLite opens in read-only mode; it never writes, repairs or prunes, though SQLite may add its usual sidecar files. |
| Clarity | 8/10 | Five fixed sections, with the two most important always printed so an empty view reads as all clear. |
| Scripting | 8/10 | --json gives the same snapshot plus a field that says which stores were read. |
| Coverage | 6/10 | Pending approvals and clarify prompts are not shown at all. |
| Multi-profile use | 6/10 | One profile per run, and the plugin must be installed in each profile. |
| Support | 4/10 | Labs tier: built by Nous Research, not maintained, no promise of fixes. |
The average comes to just over 7, which is where the headline score comes from.
I weighted nothing extra: each criterion counts the same, because a small business needs all seven to hold up.
Setup and safety score highest because the docs leave no doubt about them.
Support scores lowest for the same reason: the docs are equally clear that nobody maintains it.
What Is Actually Being Rated
Run hermes inbox and it prints a snapshot of everything the agent has open for the active profile.
Needs attention is the problem pile: sub-agent work that stalled, results lost after repeated delivery failures in the past week, scheduled jobs that broke or were paused with an explanation, dead processes left in the checkpoint, and kanban cards stuck or awaiting a reviewer.
In progress shows live work.
A third block lists delegations that finished but whose result never reached its conversation.
Scheduled lists enabled cron jobs with the next five runs, and Open surfaces lists live chat sessions with their titles.
The catalog entry lists it in the community tier and says it sends nothing anywhere.
🔥 Get the Hermes setups behind these ratings. Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members testing Hermes plugins as they land. → Join the AI Profit Boardroom
Strength: It Cannot Damage Your Setup
This is the best thing about it.
A monitoring tool that edits state is a liability, and this one is built to avoid that.
The README describes how it reads each store: the process checkpoint, the delegation table in the state database, the cron jobs file, the active-session leases and the kanban database.
The plugin's own tests check that building the view writes nothing to the home it reads.
It even works when no gateway is running, so you can check on a stopped setup.
The one caveat the README admits is minor: SQLite can create its normal -wal and -shm companion files, as it would for any reader.
That is why safety gets a 9 and not a 10.
Strength: One Command, Zero Settings
There are no model tools, no hooks and nothing that runs while you chat.
That keeps it out of the agent's prompt and out of your way.
Installation is a single hermes plugins install inbox, and the README lists no dependencies beyond Hermes itself.
I knocked one point off setup only because every extra profile needs its own install.
The agent can still use it on request: ask Hermes to run hermes inbox --json in its terminal and summarise the result.
Strength: It Says What It Could Not Read
Many status tools fail quietly.
This one lists any source it could not open under a Not read line at the bottom of the text view.
In the JSON version, a sources field records, store by store, whether it was read, missing or could not be opened.
A locked database is abandoned after two seconds rather than hanging the command.
That honesty is why I scored scripting and clarity well.
For anyone feeding a dashboard, the JSON is the same snapshot as the text view, so the two never disagree.
I held scripting at 8 because the README documents the sources field but does not publish a full schema for the rest of the output.
Limit: Approvals And Clarify Prompts Are Missing
This is the biggest gap for daily use.
A Hermes approval for a dangerous command, or a clarifying question, stays inside the memory of the process that asked it, so no outside reader can see it.
The README tells you to answer those in the chat that raised them.
So the one item most likely to block your agent is the one item the inbox cannot show.
In fairness, this is a limit of where Hermes keeps those prompts, not a bug in the plugin, and the README says so openly.
It still loses the tool points, because a reader of the view will reasonably assume it covers everything waiting on them.
Cron jobs make this sharper, because the security guide sets approvals.cron_mode to deny by default, which blocks a dangerous command in a scheduled run and leaves the agent to find another route.
Limit: Unsupported Labs Status
The banner at the top of the README is blunt: this is unsupported code from Nous.
The labs repository goes further, warning that bug reports can go unanswered and that a Hermes core update could quietly stop a plugin working, while offering each plugin to anyone willing to adopt it.
CI runs the labs plugins against the Hermes nightly build, which softens this a little.
Still, a 4 for support is fair for a version 0.1.0 tool with no owner.
Limit: One Profile At A Time
Paths resolve against the active profile's home.
To check another profile you run hermes -p work inbox, and only after installing the plugin in that profile too.
The kanban board is the odd one out: Hermes shares it across profiles, so the same card can appear in several views.
Processes inside a sandbox terminal backend cannot be probed and show as sandboxed.
How It Compares With Checking Each Store By Hand
| Task | Without the inbox | With the inbox |
|---|---|---|
| Spot failed or paused cron jobs | hermes cron list and read each job | Listed under Needs attention |
| Find undelivered delegation results | Reopen each chat | Own section |
| See next scheduled runs | Cron list | Next five runs in Scheduled |
| Find forgotten sessions | Check each window | Open surfaces |
| See pending approvals | The chat that asked | Still the chat that asked |
This comparison is my own reading of the docs; Nous does not publish one.
The cron docs remain the place to manage jobs, and our Hermes Agent cron job review covers that side.
📞 Want an expert opinion before you automate more? Book a free SEO strategy session and my team will look at your agent workflows, your content and how they connect. → Book my free session
Small-Business Examples That Fit The Score
A dental practice with overnight drafts
- A cron job drafts next-day appointment reminders into a file for reception to send.
- If the job fails, the practice manager sees it under Needs attention the next morning.
- The manager fixes the cause and reruns it.
- The inbox earns its place here, because the failure would otherwise go unseen.
A marketing agency with delegated research
- A strategist delegates competitor research to a sub-agent.
- The task shows under In progress, then under Finished, result not yet delivered if the answer never lands.
- The strategist reopens the chat and collects it.
- If the sub-agent paused for an approval, the inbox stays silent, which is where the coverage score bites.
A local bakery with a weekly order summary
- The owner schedules a Monday cron job that summarises last week's order notes into a file.
- Scheduled shows the next run, so the owner can confirm it is set for Monday.
- If the job pauses with a reason, it moves to Needs attention.
- The owner reads the reason, fixes it and resumes the job before Monday.
Here the Scheduled block does most of the work, which is why I rate clarity well even for owners who rarely open a terminal.
Who Should Use It
- Anyone running Hermes cron jobs overnight or on weekends.
- Teams that delegate research or long tasks to sub-agents.
- Operators who want a JSON snapshot for their own dashboards.
- Owners who want a read-only check they can hand to a non-technical colleague.
- Anyone who splits work across profiles and is happy to run one check per profile.
Who Should Skip It For Now
- People who mainly need to see pending approvals; the inbox cannot show them.
- Businesses that only allow supported, maintained tools.
- Anyone looking for an email inbox; Hermes handles mail through its Email gateway over IMAP and SMTP, with a dedicated account and an
EMAIL_ALLOWED_USERSlist, and outbound jobs should say "draft only".
🔥 Use this plugin with a proven routine behind it. Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video trainings and four live coaching calls a week with members who run Hermes daily. → Join the AI Profit Boardroom
Final Verdict
7 out of 10.
It is the cleanest way to see failed jobs, stalled delegations and waiting results in one place, and it is safe by construction.
It loses points for hiding approvals, for working one profile at a time, and for carrying no maintainer.
If any of those gaps get closed, or someone adopts the plugin and maintains it, the score would rise.
Until then, treat it as a reliable second pair of eyes rather than the only place you look.
It came out of core PR #81507 and reached the catalog with Hermes Agent v0.23.0, so expect it to change, or break, as core moves.
Install it, read it every morning, and keep answering approvals in the chats that raise them.
Also On Our Network
- 🌐 The six-step setup for the read-only inbox, from install to a morning cron report
- 🌐 Whether the read-only inbox is worth it for owners who leave Hermes running overnight
- 🌐 Five prompts that turn the read-only inbox into a daily Agent OS routine
- 🌐 Which models to pin to the jobs the inbox watches, with real scores
- 🌐 The complete Hermes Agent Inbox guide on the AI Profit Boardroom blog
FAQ: Hermes Agent Inbox Verdict
What score does the Hermes Agent inbox get?
7 out of 10, averaged from a seven-criterion scorecard based on the plugin README, the catalog entry and the Hermes docs.
What is its biggest weakness?
Coverage. It cannot show pending approvals or clarify prompts, because they stay in the memory of the chat process that asked.
Is it safe to run on a live setup?
Yes. It opens SQLite in read-only mode, reads JSON files as files and never writes, repairs or prunes a store.
Is it an official Nous Research plugin?
No. It is a labs plugin in the community tier: built by Nous Research, not maintained, with no promise of fixes.
Does it replace hermes cron list?
No. It flags failed and paused jobs and shows the next five runs, but you still manage jobs through Hermes cron.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Run one morning check with it, then book a free SEO strategy session or join the AI Profit Boardroom if you want help deciding where it fits next to the rest of your Hermes Agent inbox routine.
