AI Staff

Service accounts backed by an AI provider that participate in chat rooms as AI assistants.

AI staff are service accounts that join chat rooms and respond to messages using an AI provider. Each AI staff member is a regular workspace member with a name and avatar, but its replies are generated by either OpenRouter or Cursor AI instead of a human.

You can create multiple AI staff members in the same workspace — for example, one backed by a fast OpenRouter model for quick questions and another backed by Cursor for code-heavy tasks.

Prerequisites

• You must have the org.manage permission (workspace admin) to create and manage service accounts.
• At least one AI integration must be configured: OpenRouter or Cursor AI.

Creating an AI staff member

1. Open Workspace Settings

   Navigate to Workspace Settings → AI Staff (service accounts).

2. Create a service account

   Click Add AI staff and fill in:

   • Name — the display name shown in chat (e.g. Aria, CursorBot).
   • Provider — openrouter or cursor.
   • Model — for openrouter, a model ID from the OpenRouter catalog (e.g. openai/gpt-4o-mini). Not required for cursor.
   • Default repo URL (Cursor only, optional) — the GitHub repository URL the Cursor agent will use by default when none is resolved from the space’s Git links.
   • Default ref (Cursor only, optional) — the branch or commit ref to check out (e.g. main).
3. Save the token

   After creation, Hyperspeed displays a one-time service account token (sa_…). Copy it immediately — it is not shown again. You will need this token to authenticate the MCP server or other integrations.

If you do not supply a model ID for an OpenRouter service account, Hyperspeed defaults to nvidia/nemotron-3-super-120b-a12b:free. Set an explicit model to control cost and capability.

Provider options

Provider	Required fields	Optional fields
openrouter	openrouter_model	—
cursor	—	cursor_default_repo_url, cursor_default_ref

How AI staff participate in chat

Once created, an AI staff member appears as a workspace member and can be invited to any chat room. When a message is sent in a room where the staff member is present, the chat worker:

1. Gathers context — recent chat history and, for spaces with Git links, relevant file contents.
2. Calls the configured AI provider with that context.
3. Posts the provider’s response as a message from the AI staff member.

Errors from the provider are surfaced as a visible reply in the chat thread. There is no silent fallback to another model.

What AI staff can do

AI staff have the following capabilities within a workspace:

• Read chat — access recent messages in rooms they are members of (space.chat.read_recent).
• Read files — read file contents from spaces they can access (space.file.read).
• Propose patches — suggest file changes as proposals (space.file.propose_patch).

File proposals created by AI staff are not automatically applied. A human member must review and approve any proposed changes through the workspace UI.

Multiple AI staff

You can add as many AI staff members as your workspace needs. Common patterns:

• One fast, low-cost model for general Q&A.
• One high-capability model for code review.
• One Cursor-backed staff member for repo-aware coding tasks.

Each staff member has its own token, provider configuration, and room memberships. You can delete a service account at any time from Workspace Settings → AI Staff.