Skip to content

Your first chat

After onboarding, Hive opens to an empty workspace. Click the + next to Chats in the sidebar to start a new chat.

Empty chat ready for input

Type a message

The composer is at the bottom. A few affordances worth knowing:

  • Primary runtime select (above the input). Shows which runtime answers this chat, and lists every configured runtime so you can switch per chat. To address a specific agent or teammate instead, type @ to bring up the mention popover — candidates are @hive (the default agent; @primary still works as an alias), @all, your agents, and members of the chat. (There are no @file mentions; attach files with the picker below.)
  • Slash commands. Type / to bring up commands. Selecting one runs it instead of sending text: Summarize conversation, Compact conversation, Insert Linear issues, Export chat → Markdown, and Clear input — plus any user templates you've saved. /summarize and /compact manage the conversation's context window, and /linear pulls your Linear issues into the composer.
  • Attach files. Use the 📎 Attach files button in the composer to add files (including images); they're saved and referenced by path when you send — handy for "explain this" without pasting.
  • Context at a glance. The ⬡ pill in the chat header shows how much of the model's context window the next reply will use; click it for the full breakdown and controls. See Managing context.

Press Enter to send. Shift+Enter inserts a newline.

Auto-renamed titles

After the first assistant reply, Hive asks the same runtime for a 3–6 word topic summary and renames the chat. You can override at any time by clicking the pencil glyph next to the title in the chat header.

Tools

If your runtime supports tools, the model can call MCP tools you've configured. MCP servers are inert until you enable them — add and switch them on under Settings → Tools. Until then a chat runs with no tools attached.

File access for local CLI agents is governed by Claude Code's permission mode, set at Settings → Team → Agent file access:

  • default — read-only; the agent's file writes are blocked.
  • acceptEdits — the agent may create and modify files in your workspace.
  • bypassPermissions — the agent may edit files and run shell commands without asking.

Proposals — Review and Implement

When an agent proposes a change that needs sign-off, it lands as a proposal in the Review pane on the right rail. Each proposal shows its approvals (n/m), and you can approve or reject it. An approved proposal only takes effect once quorum is met and a human hits Implement — nothing is applied to the workspace behind your back. After it runs, the proposal is marked Implemented.