Your first chat¶
After onboarding, Hive opens to an empty workspace. Press ⌘N (or
the + next to Chats in the sidebar) to start a new chat.

Type a message¶
The composer is at the bottom. Two affordances worth knowing:
- Target chip (left of the input). Shows who your message goes
to. By default it's the chat's primary runtime; type
@to switch to a specific agent. - Slash commands. Type
/to bring up commands (start plan, open retrieval, attach file, switch runtime, etc.). Selecting one runs it instead of sending text. The Context group includes/summarizeand/compactfor managing the conversation's context window, and/linearto pull your Linear issues into context. - @file references. Type
@fileto reference a workspace file and pull its contents into your message — handy for "explain @src/foo.ts" without pasting. - Voice input. The 🎤 button records audio and transcribes it via a local Whisper CLI, appending the text to the composer. See Voice input & context commands.
Press Enter to send. Shift+Enter or Option+Enter inserts a newline.
Auto-renamed titles¶
After the first assistant reply, Hive asks the same runtime for a 3–5 word topic summary and renames the chat. You can override at any time by clicking the pencil glyph next to the title (or right-click the chat row → Rename).
Tool calls¶
If your runtime supports tools (Anthropic / OpenAI / OpenRouter / Claude Code / aider / pi), the model can call built-in Hive tools:
- File ops:
create_directory,create_file,propose_file_write,read_workspace_file,list_workspace_files,move_workspace_path - Shell:
run_workspace_command - Git:
git_status,git_diff,git_log,git_show,git_commit
Write tools (anything that changes the workspace) prompt you for approval inline. Approve once, for-this-chat, or always — your choice persists.

What happens after you approve¶
The tool runs in the workspace root. The result lands as a
tool_result block in the transcript so the model can read what
happened. If the tool wrote a file, the diff appears in the
Review pane on the right rail.
For details on the trust model see Tools, consent & trust.