Changelog

Regenerating or editing your chat history now requires confirmation if it will delete recent messages, preventing accidental loss of conversations. The server will also start rejecting these deep cuts after a warning period.

You can now connect multiple accounts for the same app, like home and work email, and Hermes will keep them organized. This makes it easier to manage your connections, repair them individually, and know which account was used for each task.

You can now manage multiple accounts for your connected apps like Gmail directly from the terminal interface and command line. This makes it easier to add, rename, reconnect, and remove individual accounts, giving you more control over your connections.

You can now manage multiple accounts for a single app, like having separate work and personal email addresses. This update allows you to name, rename, add, and remove each account individually within the app's connection settings, making it easier to keep track of your different logins.

A new button has been added to the toolbar that lets you easily open and close the built-in browser. This makes it quicker to access web content directly within the application.

Child processes within a conversation will now correctly send their responses back to the original conversation instead of taking over the chat. This ensures that your conversation continues as expected without interruptions from background tasks.

We've added a new test to see which AI providers actually use the reasoning information we send them. This helps us understand how well our system is working and identify which models might not be utilizing this helpful data.

We corrected a plugin version requirement for a Vietnamese language pack to ensure it works with the latest stable release. This update prevents compatibility issues and allows users to install the Vietnamese language pack without errors.

Models now remember their thought process across all interactions, ensuring a consistent and informed conversation. This improvement means your AI assistant won't forget what it was thinking about, even if you reload or switch between different ways of accessing it.

Plugins can now directly trigger built-in application actions, like showing the browser panel, using a new `ctx.runAction` command. This makes plugins more reliable by bypassing the need to simulate keyboard shortcuts, ensuring they continue to work even if users change their keybindings.

The Holographic memory provider is now a separate community-maintained plugin, making the core project smaller and easier to manage. This change happens automatically when you update Hermes, with no data loss or manual steps required.

Hermes now understands when Alibaba's content filter blocks a request, preventing unnecessary retries. Instead of waiting for minutes, users will see a clear message explaining the refusal.

Hermes now directly checks the Git branch for updates instead of first looking for a specific configuration file. This change ensures that updates on the main branch will not be blocked by network security tools that might return an error.

The holographic memory plugin has been temporarily restored to the core to ensure continued functionality until its planned removal date. This change makes no modifications to the code itself, only its location within the project.

This update fixes a bug that prevented new stable releases from being published by ensuring that older, incomplete release attempts are properly ignored. This allows the release process to continue smoothly, so you can get the latest stable versions of Hermes Agent.

The Holographic memory plugin has been moved out of the core project and into its own community-maintained catalog. This means users will now get updates for Holographic through the new catalog system, ensuring they are using a reviewed and supported version.

Hermes Agent can now open local web pages using a cloud browser. Private URLs are now securely handled by a local sidecar, preventing them from ever reaching the cloud.

Several plugins have moved to a new central location, making them easier to manage. Additionally, a desktop tool for changing accent colors is now available as a plugin.

The documentation for the Accent Picker has been updated to reflect its new home in the official plugins catalog, making it easier for users to find and install this feature. This change simplifies the installation process and ensures users are directed to the most up-to-date version.

Hermes Agent now automatically uses your local browser if the cloud browser service is unavailable, ensuring `browser_exec` continues to work without interruption. This prevents failures caused by issues like rate limits or outages with the cloud provider.

System instructions and tool configurations are now reviewed more carefully. Every change will be presented as a clear comparison, and two approvals are now required before merging to ensure important changes are thoroughly checked.

Tests for the project's code are passing again. This change updates how a specific test works to match recent code improvements, ensuring everything runs smoothly.

Tests for Hermes Agent's release process have been updated to correctly handle the new default update channel. This ensures that future releases are thoroughly tested and that the agent updates as expected in various network conditions.

The rules for approving changes to the prompt surface have been corrected to require only one core approval. This ensures the documentation and automated checks now accurately reflect the actual approval requirement.

The release process now correctly checks if the code and version tags for each built package match the expected release, preventing builds from failing at the final stage. This ensures that only accurately built and tagged packages are released.

We've removed the tabro plugin because it included a pre-built executable file that couldn't be verified. A new rule has been added to prevent plugins from including unverified compiled code in the future, ensuring that all code can be properly reviewed.

The plugin catalog now checks for and rejects packages containing hidden compiled code or unreadable files. This update makes it safer to install plugins by preventing malicious or unintended binary code from being included.

The Hermes Agent now correctly preserves the reasoning and its signature when resuming conversations or when turns are replayed. This fixes issues where reasoning was lost for Anthropic models and prevented errors when using Bedrock Converse with other models.

New configuration settings and slash commands will now fail the build if they are not accompanied by documentation. This change ensures that any new features added to Hermes are properly explained to users, preventing situations where important commands might go undocumented. Existing settings that are currently undocumented have been grandfathered in, so no previous work is affected.

You can now commit changes to documentation sidebars without the system incorrectly stopping the process. This fixes an issue where changes to documentation files were mistakenly flagged as errors, allowing for smoother contributions.

The Anatomy skill can now be installed directly within Hermes. This makes it easier to use the skill for technical explanations and geometry checks without needing external dependencies.

The installation scanner now correctly identifies fewer plugins as potentially risky. This update prevents legitimate plugins from being blocked unnecessarily, making it easier for users to install and update them.

Hermes Agent now automatically retries downloading release artifacts if their data is corrupted during the download process. This prevents build failures caused by temporary network issues or corrupted transfers, ensuring more reliable releases.

The code formatting has been automatically improved to make it cleaner and easier to read. This helps developers maintain a consistent style across the project.

Hermes Agent can now test its SSH and Docker features against live environments in our continuous integration. This means we can catch more bugs and ensure these crucial features work as expected before releasing them.

Stable releases now skip building packages for Intel Macs and Termux Android by default to speed up the release process. These can still be included if specifically requested during the release cut.

Release bundles will now be published correctly even for the very first bundled release. This fixes an issue where the release process would fail if no previous bundled release existed.

Tests now cannot use arbitrary waits that might cause flaky behavior. A new helper function is available to manage waits more reliably.

CI jobs now automatically check for and report any leftover files in the project's working directory after completing their tasks. This change helps prevent accidental inclusion of temporary or unnecessary files in releases, which has caused issues in the past.

The chat test now correctly waits for the mock model to load before sending a message. This prevents messages from going to the wrong service and ensures reliable testing.

When releasing a new version, the script now asks if you want to abandon a previous release attempt if it's stuck, making it easier to proceed with the current release. This saves you from manually stopping the old release before starting a new one.

The process for submitting verified app packages to the store now correctly checks for existing submissions. This fixes a bug that prevented new app versions from being published.

The ability to turn off shell persistence for SSH and local environments is now functional. This means that if you previously disabled persistence, each command will now start fresh, ensuring better isolation between them.