Changelog

Connecting apps is now handled entirely by Hermes' backend, making the connection process look and feel the same across the desktop app, terminal, and command-line interfaces. This unified experience simplifies how you add and manage app connections.

A new plugin, Cron Bulletin, has been added, allowing Hermes Desktop users to view and manage their local cron reports directly within the application. This makes it easier to track and control scheduled tasks without needing to interact with system files.

The CDP Manager plugin has been updated to version 1.0.2, which fixes an issue where Chrome could get stuck in a profile picker screen, preventing new tabs from opening. This update ensures that new browser tabs can be reliably created and adds new actions to detect and fix this problem if it occurs.

Settings for custom AI endpoints are now correctly linked to your chosen profile, making it easier to manage different API keys without extra steps. This ensures that your custom endpoints and API keys are saved to the right place and appear immediately in your model selection.

The plugin catalog now uses the latest version of the bridge adapter, which includes improved documentation for installing and using the SDK. This change makes it easier for users to understand and set up the adapter.

The tool that automatically creates titles for your content has been updated to its latest version. This improves the reliability and accuracy of title generation.

The RSS Reader plugin has been updated to version 1.0.7. This update improves how you subscribe to content, manage individual feeds, and view your subscribed items.

The Hermes Agent can now discover and utilize new tools through the ard-kit plugin. This integration allows agents to find and use resources for tasks like rendering web pages or sending SMS messages directly from a local catalog.

The OpenMail plugin has been updated to a newer version, improving security by no longer requiring the agent to create a special inbox key. This change makes the plugin safer to use and fixes issues with sending attachments.

The `hermes backup` command will now exclude browser user data, making your backups smaller and faster. This prevents sensitive login information from being accidentally included in your archives.

The configuration loading process now correctly handles numeric model provider IDs, preventing the gateway from crashing. This ensures that settings like `provider: 2` are interpreted as text, not numbers, allowing the agent to function properly.

The `hermes doctor` command now correctly signals when it finds unresolved issues by exiting with an error code. This allows automated checks and monitoring systems to accurately detect and report when problems remain unfixed.

The system now correctly identifies when a running gateway is using outdated code and displays this as a problem, prompting a restart. This ensures gateways are always updated to the latest version, preventing potential issues caused by stale code.

The Hermes Agent can now include a new community plugin called "kanban-task-threads." This plugin helps organize tasks by creating a dedicated discussion thread for each one, making it easier to track progress and conversations.

Parked directories that aren't valid profiles no longer show up as teammates in bot mode. This makes the list of available teammates more accurate and easier to use.

When requesting a review for a card that's already assigned to its reviewer, Hermes now correctly avoids assigning the reviewer as the implementer. This prevents review feedback from being incorrectly sent back to the person who originally requested the review.

The `hermes kanban show --json` command now correctly displays the maximum runtime allowed for a task. This helps operators easily check runtime limits directly from the command line.

The Hermes CLI will now safely prevent sensitive information like API keys from being accidentally exposed when running shell commands if a security tool fails to load. This makes the command execution more secure by ensuring it either runs correctly with proper isolation or fails clearly without leaking data.

The Hermes Agent no longer hangs when interacting with certain OAuth-protected services. This is because it now correctly identifies and only processes authorization server discovery documents, avoiding issues with long-running data streams.

Important settings for how Hermes behaves are now correctly applied to each specific profile being used, instead of using a single setting for the entire application. This ensures that different profiles can have their own unique behaviors and configurations.

Login attempts with expired refresh tokens will now correctly indicate that the login is invalid and require re-authentication. Additionally, using login aliases will now store credentials under their primary provider name, ensuring consistent status checks and runtime behavior.

When a configuration file is missing the setting for compression token limits, the agent now correctly uses the default limit instead of removing it entirely. This ensures that the compression process stays within reasonable bounds and prevents excessive memory usage when handling large models.

Connections to Google's services like Gmail and Calendar will now stay active longer and can be reused easily. The device login flow for Google has also been fixed to prevent errors.

Hermes now remembers profile information between updates, making it much faster to view your agent's status. This means your agent will feel more responsive, especially when you have many profiles or sessions.

The way Hermes handles commands for the browser has been reorganized to make it easier to add new commands in the future. This change also improves how some browser commands are processed, preventing unexpected behavior with slightly different inputs.

The agent now waits for the exact amount of time specified by the provider when it's temporarily blocked, instead of guessing. This means the agent will try again sooner when the provider is ready, improving reliability.

The Photon sidecar's watchdog can now correctly detect if a stream is no longer active by using a unique identifier for its test messages. This ensures that Hermes can properly identify and handle unhealthy connections.

When Hermes Agent sends a reply through a streaming connection, the sender now correctly receives that reply text. This ensures that all relevant Hermes Agent endpoints show the actual reply, not just an empty confirmation.

Previously, messages sent during a conversation could appear multiple times in the chat history. This update ensures that these messages are now displayed only once, making the transcript cleaner and easier to read.

Group chats no longer lose messages when a hold is applied, and you can now turn off automatic message holding in specific rooms. This means important messages won't be accidentally skipped, and you have more control over how your group chats behave.

Non-administrator users will now only see the commands they are allowed to use when asking for help, preventing confusion and unnecessary error messages. This makes it clearer for users to understand what actions they can take within the system.

Messages sent to platforms that don't support streaming or editing will now be sent only once, preventing duplicates. This change ensures a more reliable experience for users on platforms like Weixin and QQ.

A new community plugin, "prompt-tray," has been added, allowing you to easily access and insert saved chat prompts directly into the composer. This feature helps streamline your workflow by providing quick access to pre-written prompts for common tasks like reviewing, planning, or explaining.

The warning about your token being stored in plain text will now only appear if your computer can't securely store it, and the advice given will be more accurate for your operating system. This prevents unnecessary warnings for users who have chosen not to use extra security features.

Hermes Agent now more accurately identifies which browser processes belong to a specific session, preventing accidental termination of unrelated processes. This improves stability by ensuring only the correct browser processes are cleaned up.

When managing dashboards that are set up without a specific home directory, Hermes will now correctly associate them with their owner's home. This ensures dashboards are properly tracked, even when running in minimal environments.

The core logic for starting up and running the text-based interface has been organized into separate helper files. This makes the main code file shorter and easier to manage without changing how Hermes Agent works.

Hermes now correctly launches the user interface specified in your profile's configuration file when you use the `-p` option. This ensures that Hermes starts up with the correct settings you intended for that specific profile, rather than defaulting to the main configuration.

Pausing or stopping voice detection is now faster and more reliable, even if the microphone stops working. This prevents the agent from freezing and ensures a cleaner shutdown.

The selection of DeepSeek models is now more accurate, showing only the latest available versions. This makes it easier to choose the correct model by displaying clearer names like "DeepSeek V4.1 Flash".

The name shown for models in the chat composer now correctly displays the user-friendly name of the provider, like "xAI Grok," instead of technical internal identifiers. This makes it clearer which service is being used to power the AI model you're interacting with.

The Kanban board now prevents you from marking a task as complete if you haven't provided any details or a summary, ensuring that all completed tasks have some record of their outcome. This change improves tracking by requiring a result or summary for every completed task.

The cron job editor now correctly saves custom provider names, preventing them from being misinterpreted and ensuring they are used as intended. This fixes an issue where custom model configurations were being lost or causing errors.

When you disconnect or change a desktop connection, Hermes now properly forgets its old information. This ensures that if you reconnect or point it to a new location, it correctly learns your profile and identity from the new source instead of using outdated details.

SSH connections now correctly identify the same machine, so duplicate entries for the same device will appear as a single item in your list. This makes managing your devices cleaner and easier.

When the application's window unexpectedly closes due to system issues, it now displays a helpful recovery page explaining the problem and offering a reload option, rather than just a blank screen. This change makes it clearer to users why the window closed and how they can try to reopen it.

When a process managing tasks dies unexpectedly, it now includes the last few lines of output and a summary of the code's current state. This helps the system recover and continue its work more effectively without needing to manually inspect logs.

User-added authentication providers now correctly use their own names and aliases when registering, ensuring that custom configurations take precedence. This fixes an issue where built-in Hermes settings could unintentionally override custom provider details.

The agent now better manages how many AI models can run on your computer's graphics card. It will avoid loading a new model if there isn't enough memory, preventing performance slowdowns.

The `hermes update --list-venv-holders` command now outputs a list of running processes that might prevent an update in a structured JSON format. This allows automated updates to precisely identify and stop these processes, preventing update loops and ensuring a smoother upgrade experience.

The right side of the desktop app now correctly shows only the preview tabs associated with the chat you're currently looking at. This means previews from one agent's chat won't clutter the views of other agents.

When Hermes Desktop updates itself from a remote source, it will no longer unnecessarily restart its local messaging gateway. This prevents conflicts and ensures a smoother update process.

The prompt snippet manager has been updated with a cleaner interface and now allows you to easily enable or disable individual snippets. This makes managing your prompts more flexible and user-friendly.

Private Google Workspace links that lead to a sign-in page will no longer unexpectedly trigger your operating system's passkey prompt. The app now correctly identifies sign-in pages without needing to open a hidden browser window, keeping the link preview focused on the actual content or a placeholder.

The update process now correctly identifies when your gateways are already running the latest code, preventing unnecessary restarts and a confusing "no gateway running" message for Windows users. It also ensures that incomplete update receipts are properly resolved, avoiding error messages after a manual restart.

You can now edit messages in a conversation even after jumping far back in the chat history. This fix ensures the editing feature remains available and functional, no matter how far you scroll.

When the automatic summarization model fails, Hermes Agent will now try the main summarization model as a backup. This change also improves error messages by clearly indicating which summarization model caused the problem.

The chat composer on desktop will now remain visible during brief loading periods, preventing the application from crashing when you're typing. This change ensures a more stable and consistent chat experience.

Pasted text is now correctly attached and processed by the agent, even when it's not in the current project directory. This ensures that large pasted messages are not lost and can be used by the agent.

The code for managing systemd services has been moved into its own file. This change helps keep the main gateway file more organized and makes it easier to understand and maintain the service management features.

The Hermes Agent now correctly prevents a stale gateway from incorrectly dispatching jobs after an update. This ensures that only truly fresh gateways take over tasks, improving stability and reliability.

Language servers will now recover more gracefully if they encounter an error. A single stalled language server will no longer prevent other parts of your workspace from functioning, and you can now configure specific directories to be excluded from language server processing.

Hermes Desktop now frees up memory by releasing old chat messages from parked conversations, and it better manages the memory used by rendering processes and agent avatars. This makes the application more stable and efficient, especially for longer usage sessions.

This update fixes an issue where messages could be sent twice or not at all. Now, the system correctly tracks delivered messages, ensuring each message is sent only once and that refused messages are still handled.

You can now safely switch your Hermes database between WAL and DELETE modes using a new command, `hermes sessions set-journal-mode delete|wal`. This command ensures the database is not in use and verifies the switch, preventing data corruption and providing clearer error messages.

Messages typed into a chat now go to the correct conversation even if that chat is being compressed in the background. This prevents unexpected behavior where typing in one chat would send messages to another.

Storage maintenance commands will now refuse to run if another Hermes process is actively using the database, preventing potential data corruption. If this happens, the agent will provide clear instructions on how to resolve the issue and offer a link to a recovery guide.

The main `cli.py` file has been significantly shortened by breaking down its functionality into smaller, specialized modules. This makes the code easier to manage and understand, improving the overall maintainability of the project without changing how it works.

The Hermes Agent now stops unnecessary background animations when they are not actively needed, which reduces its usage of computer resources. This change makes the application more efficient and prevents it from consuming power and processing time when idle.

The Hermes Agent can now discover and use zero-cost AI models from a new provider, making it easier to access a wider range of tools. This update also includes detailed documentation on how these models can work with Hermes.

A new adapter for the Orbit Desktop browser workspace has been added, allowing Hermes Agent to interact with your local Orbit sessions. This integration provides enhanced control over your workspace settings and data.

The Claude plugin now works on Windows, not just Linux and macOS. The plugin's catalog entry also displays its associated artwork.

The Hermes Agent now prevents multiple processes from writing to the same database simultaneously, ensuring data integrity. It also improves recovery by allowing restored databases to function correctly on the first attempt and stops imports from overwriting active databases that have been deleted.

The plugin catalog now shows when each plugin was added and last updated. You can also sort the plugins by when they were added or recently updated, making it easier to discover new and changed tools.

The plugin catalog now features dedicated pages for each plugin and author, making it easier to share and discover information. These new pages include optional screenshots and the plugin's README, providing a richer experience for users.

We fixed a test that was incorrectly flagging a valid command as an error. This change ensures that the test accurately reflects how Hermes handles different session commands.

This update improves crash reporting on Linux by ensuring that detailed log messages are saved when an error occurs, even if the main application log is otherwise silent. This helps developers understand the cause of crashes by providing more diagnostic information locally.

You can now uninstall plugins directly from the Hermes Desktop Plugins hub, making it easier to manage your installed extensions. This feature addresses a common request from users who wanted a simpler way to remove plugins without resorting to the command line.

Subagents now correctly remember their specific provider identity when inheriting it from a parent agent. This ensures that specialized tasks like analyzing vision data use the most efficient method and that subagents access the correct shared resources.

The Hermes Agent no longer sends your username, home directory, or current location to the model when probing remote environments. This improves privacy by only sharing necessary operating system information.

Hermes Agent now limits the number of MCP servers it starts simultaneously during boot. This prevents a flood of server processes from overwhelming the system and improves startup stability when many servers are configured.

The way Hermes Agent remembers conversations has been improved to work better with smaller memory limits. Now, it will always leave enough room to summarize, ensuring that the agent can effectively compact conversations even for less powerful models.

RoomLink catalogs now correctly identify the profile they are associated with and use that profile's specific settings for their execution policies. This ensures that each catalog uses the intended configuration, preventing unexpected behavior when multiple profiles are in use.

The assistant's chat history now correctly updates as new messages are added, and the count of messages shown when jumping through history is now accurate and consistent. This ensures a smoother and more reliable experience when reviewing past conversations.

Failed group chat members now show a more specific error message, making it easier to understand why a member's turn failed. This helps diagnose issues with the backend or provider.

The Hermes Agent's command line interface has been reorganized to make it more manageable. The code for handling the launchd backend and the setup wizard has been moved into separate files, improving clarity without changing how the agent functions. This makes the project easier to maintain and understand.

Plugins in the catalog are now easier to explore. Clicking on any plugin card will directly take you to its dedicated page, and more information about each plugin is always visible.

Hermes Desktop now correctly attaches to an already running backend on your machine instead of starting a new one. This ensures only one backend runs per host, saving resources and preventing conflicts.

The way memory replacements are handled has been improved to ensure that entire entries are preserved and accounted for, preventing accidental data loss. Now, when you replace a piece of memory, the system will clearly show you what was overwritten, making it impossible for the change to be silently incomplete.

Plugin pages now automatically display their documentation, making it easier for users to understand what each plugin does. This change ensures that all plugin pages have visible README files, improving discoverability and usability.

Hermes now prevents multiple instances of the same command from running simultaneously on your machine, ensuring only one instance manages your profiles. If you try to start a second instance, it will now safely connect to the existing one instead of causing conflicts.

Hermes now correctly cleans up and manages the database files for secondary profiles. This ensures that these files don't grow indefinitely and that retention settings are applied properly, preventing potential performance issues and data inconsistencies.

The Hermes Agent now smartly trims older messages from chat history, keeping only what's needed for the visible window and a bit of extra. This change significantly reduces memory usage, making the application run more smoothly by preventing unnecessary storage of past conversation details.

When your computer's settings for Hermes show it has permission but it still doesn't work on macOS, the Hermes Agent will now explain how to fix it. This addresses a problem where software updates could cause the system's permission settings to become outdated and unresponsive.

When multiple profiles are running, Hermes Agent now correctly uses each profile's own settings instead of the main launch profile's. This prevents sensitive information from being exposed to the wrong profile, making the system more secure.

The system now correctly manages background tasks for each profile independently, preventing conflicts and ensuring tasks run according to their specific configurations. This makes the system more reliable by ensuring each profile's scheduled jobs are handled properly.

The Hermes Agent no longer requests an unnecessary permission related to invoking tools. This simplifies how the agent requests permissions and ensures it only asks for what it truly needs.

The Hermes Agent now accurately shows a single host gateway serving multiple profiles, instead of incorrectly reporting that a profile's gateway is "not running." This change makes it clearer when your gateway is functioning correctly and avoids unnecessary confusion.

The Hermes Agent now ensures only one gateway process runs on your machine, managing all your profiles. This change simplifies how commands interact with the gateway, preventing conflicts and ensuring a more stable experience.

Hermes Agent on macOS can now connect to devices on your local network. This fixes an issue where the agent, when running in the background, was unable to reach devices like printers or smart home hubs.