Skip to main content
The Chat tab is the heart of Herm. It occupies the left pane and is the first thing you see when you launch herm. Every conversation with Hermes Agent flows through this interface: you type in the composer at the bottom, responses stream in above it with full markdown rendering, and the sidebar alongside it shows context window state and model details.

​
Streaming responses

Herm renders each response as it arrives. You don’t wait for the full reply before reading — tokens stream in and the display updates continuously. The renderer handles:
  • Markdown — headings, lists, bold, code blocks, blockquotes, and tables all render natively in the terminal.
  • LaTeX-to-unicode — inline and block math expressions are converted to unicode characters so formulae are readable without a browser.
  • Inline images — when Hermes Agent produces or references an image and you have chafa installed, the image renders inline inside the chat pane.
If images appear as broken placeholders or raw escape sequences, install chafa. On macOS: brew install chafa. On Debian/Ubuntu: apt install chafa.

​
Diff chips

When Hermes Agent edits files during a task, Herm surfaces each change as a diff chip — a compact, tabbed frame in the message stream that shows exactly what was added or removed. You can navigate the chip’s tabs to review each changed file individually without leaving the terminal.

​
Expandable tool calls

Tool invocations are shown inline in the message stream as collapsed entries. Press Enter (or click) on any tool call entry to expand it and inspect the full input and output. This lets you follow exactly what the agent is doing at each step without the raw JSON overwhelming the conversation.

​
Thought cloud

For reasoning models that expose their thinking steps, Herm shows a thought cloud above the final response. This collapsible block contains the model’s intermediate reasoning so you can audit its logic before it commits to an answer.

​
Composer

The composer sits at the bottom of the Chat tab. The status dot to the left indicates readiness: green means ready, yellow means generating, red means disconnected.
KeyAction
EnterSend message
Shift+EnterInsert a newline (multi-line message)
Ctrl+VAttach a clipboard image
↑ / ↓Navigate prompt history
/Open the slash command popover
@Open the file/diff reference picker
While a response is streaming you can still type in the composer. Your input queues automatically. Press Enter to add it to the queue; it sends as soon as the current turn finishes. The composer placeholder changes to “Type to queue… (Enter queues, click chip to edit)” to remind you of this.

​
@ references — attaching file and diff context

Type @ anywhere in the composer to open the @ reference picker. This lets you attach local files or diffs directly to your message so Hermes Agent has precise context without you pasting content manually.
1

Type @ in the composer

A popover appears listing files and diffs that match the text you type after @. Use ↑ / ↓ to navigate the list.
2

Select a file or diff

Press Enter to confirm the selection. It is inserted as a styled chip in the composer. You can attach multiple references in a single message.
3

Send your message

Press Enter to send. Herm expands the chips and forwards the referenced content as context alongside your prompt.
You can also paste a file path directly into the composer. Herm detects it, stats the file via the gateway, and — if it is an image — attaches it as a preview chip automatically. Pastes of five or more lines are collapsed into a [Pasted #N …] placeholder that Hermes Agent expands server-side.

​
Voice mode

Voice mode lets you speak prompts instead of typing them and optionally have responses read back aloud. 
/voice on       # Start recording
/voice off      # Stop recording
/voice tts      # Toggle text-to-speech readback
/voice status   # Show current voice state

/voice on

Begin recording. Herm captures audio and transcribes it into the composer when you stop.

/voice off

Stop recording and submit the transcribed text.

/voice tts

Toggle text-to-speech. When enabled, agent replies are read aloud as they stream in.

/voice status

Print the current recording and TTS state to the chat.

​
Useful slash commands in Chat

Type / in the composer to open the slash popover and browse all available commands. A few that are especially useful while chatting:
CommandDescription
/newStart a fresh session
/clearClear the visible chat messages
/steer [text]Inject a steering note mid-turn without interrupting the agent
/modelSwitch the active model
/themeBrowse built-in color themes
/keysView and rebind keyboard shortcuts
/queue [text]Queue a prompt for the next idle turn
/stashPark the current prompt draft for later
/stash popRestore the most recently stashed draft
Press Ctrl+K at any time to open the command palette for a searchable view of all actions.