SecuFocus
BriefAnnouncements

Hermes Agent: one shared gateway, profiles to check separately

The 0.22 cycle brings profiles into a shared gateway and Desktop backend. Check which profiles are served before starting a second instance or stopping the host.

Generated image: a black network switch with three differently coloured cables on a wooden desk.

Generated image · SecuFocus

At a glance

Key points

In the 0.22 cycle, Hermes can serve multiple profiles from one messaging gateway and one Desktop backend. There are fewer processes to maintain, but stopping the host is not the same as stopping one bot. After migration, check the profiles actually being served and the accounts they use.

An announced migration still needs a status check

The 0.22.0 release makes a single gateway the default and announces migration of older per-profile setups during an update. The documentation adds safety checks: an old gateway still running, duplicate bot credentials or an incompatible platform can block the automatic fold.

After updating, read hermes gateway status. On a multi-profile machine, a successful start of the default profile alone does not prove the others are served. A refusal should identify unserved profiles and the blocker. Do not start another gateway blindly to make the warning disappear.

Stopping one profile does not require stopping the host

In the documented shared mode, gateway stop applied to a named profile parks its adapters and excludes its scheduled jobs from subsequent ticks. It deletes neither sessions nor saved jobs. Profile start and restart commands can bring it back without stopping the host.

This does not cancel work already dispatched by a scheduled job. Profiles are not separate machines either: the release describes profile-specific configuration, allowlists and sessions, but also notes that a root OAuth login can remain shared when a profile has no login of its own. Check the actual account, not just the profile name.

Telegram: a message in the lobby may continue the latest topic

Release 0.22.3 announces opt-in Telegram DM topic transfer. The documentation explains that, with /topic mode enabled, an ordinary message sent in the root DM continues the most recently active topic by default. It does not necessarily start a new conversation.

The dm_topic_auto_transfer setting, off by default, requests a new topic for each ordinary root-DM message instead. Slash commands do not create topics. If Telegram refuses topic creation, Hermes falls back to the most recently active topic. Check where the reply actually appears before changing the project or account in your request. Within a topic, /new resets that session without resetting the others.

Messaging changes do not grant the same access everywhere

The 0.22.0 cycle adds a native /model picker in Slack and uses its Agent Sessions API for status and titles. WhatsApp can keep a separate paired session for each profile. Its documentation says an enabled but unpaired profile is skipped with whatsapp_unpaired status: enabling a profile does not prove it can receive or send through that account.

The same release announces monospace tables in Signal, LaTeX rendering in Matrix and media delivery through send_message for QQ Bot. For Element, the documentation requires enabling maths rendering in Labs; another client may show raw text. These display and delivery improvements do not by themselves change who may instruct the bot. Review each platform’s DM and group access policies before inviting it into a room.

Button language does not set the language of replies

The 0.22 cycle introduces shared translation packs across the CLI, gateway, TUI and Desktop. A profile can also override individual labels with its own translation file.

The documentation limits this to static interface text. Model replies, tool results, logs and command names are not translated by that setting alone. Ask the model separately for your preferred reply language. An incomplete pack can leave some labels in English.

Visible work is not necessarily durable work

The 0.22.0 release describes live subagent monitoring with tasks, elapsed time, transcripts and controls to steer or stop them. The 0.22.3 release adds steering and editing queued prompts in Desktop, as well as live progress from MCP calls.

Delegation remains tied to the owning session and process. Closing the session, using /stop or /new, or restarting Hermes can interrupt the work. For an operation that must survive those events, choose a durable mechanism rather than assuming the activity panel will resume it.

These are control improvements described by Nous. SecuFocus has not migrated an installation or tested a multi-profile shutdown for this article.

Check and explore

Sources for this article

Numbers connect each reference to the passages that use it. Dates show when the documentation was consulted.

This article draws on the sources above. The exercises are for you to try on your devices; SecuFocus does not present them as tests carried out by its editorial team. Interfaces and features can change. Method and corrections.

Cite this article

Keep this reference with the article when you save or share it.