Hermes Releases 105 Plugins Installable With One Command

Hermes Agent’s new plugin catalog drops 105 ready-to-install extensions in a single command, giving developers a faster way to add functionality while promising tighter security and better memory handling.

The catalog, unveiled by Nous Research, bundles four official extensions with 101 community contributions across nine categories. Most of the tools sit in the “Tools” and “Desktop” sections, reflecting the everyday workflows of AI-augmented users.

Why the catalog matters now

AI agents often stumble when a session ends and they lose context, forcing users to repeat information. At the same time, the open-source nature of many extensions raises concerns about malicious code slipping into a developer’s environment. By combining a one-command install flow with SHA pinning and a mandatory review step, Hermes tackles both problems in a single package.

How the security model works

The Hermes team signs every plugin with a SHA hash—a cryptographic fingerprint that ties the package to a specific version. When a user runs the install command, the system checks the hash against the recorded value, confirming the code hasn’t been altered after publication. The team also screens each community submission before it appears in the index, adding a human layer of verification.

The safety net doesn’t end there. Project-level plugins, which can run arbitrary code from cloned repositories, start disabled. Users must flip a switch to enable them, reducing the chance of accidental execution of untrusted scripts.

Memory-focused extensions

A recurring complaint among Hermes users is the agent’s tendency to “forget” data between sessions, inflating token usage and slowing down interactions. The catalog’s Memory category highlights plugins designed to persist context or compress chat logs. For example, hermes-snapcompact turns conversation histories into PNG images, cutting down the number of tokens the model must process. The jackal-verified plugin adds a verification layer for 41 different tools, helping agents trust the data they retrieve.

The web-search-plus plugin leads the community with 402 stars, indicating strong demand for integrated browsing capabilities. Security-focused developers may gravitate toward the official snyk plugin, which scans code and library dependencies, pins versions, and disables analytics to lower the attack surface.

The hermes-telegram-business extension turns the agent into a secretary bot, but it forces a manual approval step for every outbound message—an extra safeguard for client-facing communications.

Using the catalog

Hermes provides a tiny CLI to manage extensions:

  • hermes plugins – launches an interactive screen for browsing.
  • hermes plugins list – prints a table of installed plugins.
  • hermes plugins search [name] – looks up a specific entry in the index.

Because the entire catalog can be fetched with a single command, teams can spin up a fully equipped environment in minutes rather than hours of manual setup.

Potential downsides

Even with SHA pinning and review, community plugins remain a vector for supply-chain risk. A reviewer might miss subtle backdoors, and the hash verification only guarantees the code matches the vetted version—not that the version is safe in all contexts. Users must still assess whether a plugin’s functionality aligns with their security policies.

Memory-saving tricks like image-based chat compression can reduce token counts, but they also strip away searchable text, possibly limiting downstream analysis. Organizations should weigh the trade-off between efficiency and data accessibility.

What to watch next

If the catalog gains traction, more categories will likely appear, and a tiered trust system could let plugins earn higher visibility after longer periods of clean operation. Monitoring how often the community enables project-level plugins will also reveal whether the default-off stance effectively curbs accidental exposure.

Takeaway: Hermes’ one-command, 105-plugin catalog offers a streamlined path to richer AI agents, but its security hinges on diligent review and user vigilance—especially when extending beyond the officially vetted set.