The Hermes credential vault is the new password-blind login system that shipped in Hermes Agent v0.21.2 on 11 September 2026: your agent can now sign in to websites, pay for things and fill in addresses using credentials pulled from 1Password or Bitwarden, and it does all of that without the model ever viewing the secrets themselves, according to the official v0.21.2 release notes published on GitHub by Nous Research. That single change removes the ugliest compromise in agent automation — the one where you had to paste a real password somewhere an AI model could read it — and it is the standout feature in a release otherwise focused on database reliability and profile isolation.
📺 Watch: Hermes Agent Can Now Browse the Web as YOU
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside · Want AI SEO help 1-on-1? Book a free SEO strategy session →
What the Hermes credential vault actually does
Before this release, letting a Hermes agent log in to anything meant handing it the password directly — typed into a prompt, dropped into a config file, or stored somewhere the model could read on demand. Every one of those paths puts a live secret inside the model's context, where it can end up in a transcript, a log file or a session export. The Hermes credential vault closes that gap. Per the v0.21.2 release notes, the agent can now "sign in, pay, and fill addresses" using your existing 1Password or Bitwarden vault while staying password-blind throughout: the secret material is handled by the password manager integration, not passed through the model.
The distinction is worth sitting with for a second. A credential the model never sees is a credential that cannot leak through the model. The agent gets the outcome of authentication — a logged-in session, a completed checkout, a filled form — without ever holding the raw password that produced it. That is the same principle good human security practice has pushed for years: you use a password manager precisely so the secret never sits in a place it does not need to be.
Why password-blind logins change what your agent can do
Most of the genuinely valuable jobs you would want an agent to run sit behind a login wall. Checking a supplier dashboard, pulling a report from an ad platform, renewing a domain, updating a listing, buying a tool licence — an agent without credentials has to stop at the sign-in screen and hand the job back to you, which defeats the point of delegating it. With the credential vault, those workflows become finishable end to end. If you want a sense of what agents are already doing for people day to day, the Hermes agent use cases round-up covers the popular ones, and the video above shows the browse-as-you behaviour this vault now makes safer.
It also pairs naturally with the rest of the stack. The vault handles who your agent can act as; a proper operating layer decides what it should actually do with that access. The Agent OS guide covers how to structure agent work into repeatable systems, and connecting outside tools through the Hermes MCP server is how most people wire those systems into real services.
If you want working agent systems doing real business tasks for you every day — not just demos — check out the AI Profit Boardroom → join the community here. Prefer to map your own setup out 1-on-1 first? Book a free SEO strategy session and get a plan before you build.
How to get the Hermes credential vault
Type hermes update in any terminal. That is the official upgrade path for existing installs, and the v0.21.2 release notes document no breaking changes, so it is a low-drama update. The full walkthrough — including what to check afterwards — is in the how to update Hermes Agent guide, which covers the same 0.21.x update flow. If you have never installed Hermes at all, start with the Hermes agent installation guide instead and you will land on the current version from the first install.
Which password managers does the vault support?
The release notes name two: 1Password and Bitwarden. Nothing else is listed in the v0.21.2 announcement, so if you keep your credentials in a different manager, the safe assumption for now is that you would need to move the specific logins you want your agent using into one of those two, or wait to see whether Nous Research extends support in a later release. Given that Bitwarden has a free tier, there is a workable route for the vault regardless of what you currently pay for.
📺 Watch: 5 FREE Hermes Agent Use Cases That Are Actually Insane
What else shipped in Hermes Agent v0.21.2
The credential vault is the headline, but v0.21.2 is mostly a reliability release, and the fixes matter if you run agents daily. From the official release notes:
- Database reliability: the state.db campaign eliminated the concurrent writers that could corrupt databases, moved hosted rooms to a shared-state.db, and fixed a sticky error that flagged healthy databases as corrupted. Full-text-search damage no longer crashes a conversation, and a single malformed row no longer kills your session listings.
- Multi-profile isolation: secondary profiles no longer inherit the default profile's allowlists, credentials no longer leak to unintended profiles, and webhooks now route to the correct profile. If you separate client work into profiles, this is the fix you wanted.
- Desktop performance: the spawn storms that hit when switching profiles or hovering the roster are gone.
- Plugins: there is now a curated plugin catalog with SHA pinning and a unified Plugins page for both agent and desktop. If you are choosing what to install, the best Hermes agent plugins list is a good starting shortlist.
- Free tier and onboarding: free Nous inference and connectors, a one-command signin with /login from chat, and a guided first-launch mode for new users.
Is the Hermes credential vault safe to use?
Password-blind design is a genuinely stronger model than anything the community was doing before — pasting secrets into prompts was always the weakest link in agent automation. But be clear-eyed about what you are granting: an agent that can sign in and pay is an agent with real authority over your accounts and your money. The sensible approach is to scope it. Give the vault access to the specific logins an agent actually needs for its job, not your entire password manager, and keep payment methods limited to cards with sensible caps. The multi-profile isolation fixes in this same release exist precisely because credentials crossing boundaries they should not is a real failure mode — v0.21.2 tightening that at the same time it introduces the vault is not a coincidence, it is the two halves of one security story.
Hermes credential vault vs pasting passwords into your agent
| Credential vault (v0.21.2) | Pasting passwords manually | |
|---|---|---|
| Where the secret lives | Inside 1Password or Bitwarden | In prompts, configs or memory files |
| Can the model read it? | No — password-blind by design | Yes, in full |
| Leak surface | Password manager only | Transcripts, logs, exports, screenshots |
| Revoking access | Change it in the vault once | Hunt down every place you pasted it |
| Payments and addresses | Filled without exposure | Typed into context |
The comparison is one-sided because it should be. There is no workflow where the old approach wins; the vault is simply the correct way to let a Hermes agent authenticate from now on.
Where the credential vault fits in your bigger Hermes setup
Think of v0.21.2 as the release that made Hermes agents trustworthy with real accounts. Pair the vault with a tidy working environment — the Hermes desktop app is the easiest way to watch what your agent is doing while it acts on your behalf — and pick the model brain deliberately rather than by default: the Goldie Bench write-up covers how the different brains you can run inside Hermes compare in hands-on tests, which is exactly the question that matters once an agent is handling logged-in work. From there, the pattern is the same as always: one agent, one job, one scoped set of credentials, and expand only when each piece runs clean.
If you want the full Agent OS, prompt libraries, daily tutorials and weekly live coaching calls to put agents like this to work in your business, check out the AI Profit Boardroom → get inside here. And if you would like a personal roadmap before you commit, book a free SEO strategy session — it costs nothing and you leave with a plan.
Real wins from inside the AI Profit Boardroom
See all 3,000+ members →
Ready To Join The #1 AI Community?
Join 3,600+ entrepreneurs inside the AI Profit Boardroom. Get 1,000+ plug-and-play AI agent workflows, daily coaching, and a community that holds you accountable.
Join The AI Community →7-Day No-Questions Refund • Cancel Anytime
Wait — before you leave
Get the entire AI Profit Stack for $59/mo locked
- 🎯 1,000+ done-for-you AI agent workflows
- 📅 5 live coaching calls per week with me
- 👥 3,000+ AI operators inside
- 🛡️ 7-day refund + 30-day ROI guarantee