> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hyperspell.com/llms.txt
> Use this file to discover all available pages before exploring further.
# How agents use Hyperspell
> What an AI agent can do once it is connected to your company brain, and the ways to connect one.
An AI agent on its own knows only what is in the current conversation. However just like humans, agents produce their best work if they have context on the goals, decisions, history, and future plans. Hyperspell provides that context to agents.
When you connect an agent to Hyperspell, it can query the company brain to get grounded answers, with a paper trail of where the information came from.
## System Prompts
We recommend using an additional system prompt to guide the agent's behavior.
Hyperspell is the source of truth for everything about this organization: people, roles, projects, decisions, history, and priorities.
When starting a new task, use the Hyperspell MCP to get the context on the task, goals, prior decisions, and how it matters to the organization. Before answering any question that touches the organization — or making an assumption about who someone is, who owns what, or the state of a project — consult the Hyperspell MCP. Never guess about people, ownership, or project state when Hyperspell can tell you.
You can add this to your Instructions for Claude in the personal settings. Admins can also add this to the Organization instructions at [claude.ai](https://claude.ai/admin-settings/organization).
Additionally, you can add your agent's work, decisions, and insights into the brain to save it for later.
Hyperspell is the source of truth for everything about this organization: people, roles, projects, decisions, history, and priorities.
When starting a new task, use the Hyperspell MCP to get the context on the task, goals, prior decisions, and how it matters to the organization. Before answering any question that touches the organization — or making an assumption about who someone is, who owns what, or the state of a project — consult the Hyperspell MCP. Never guess about people, ownership, or project state when Hyperspell can tell you.
Before compacting, or when done with a task, use the `remember` tool to save your decisions to the brain. Things that should be remembered:
* Decisions made and summaries of tasks you completed
* Complex tool calls that took several reasoning steps to figure out the right tools and arguments
* Insights about the user, their work, and the organization
## Agents act as you
Every agent you connect acts with your identity and your permissions. It sees the documents you can see: everything shared with the organization plus your own personal data. It cannot reach a colleague's private inbox any more than you can. See [Permissions](/core/permissions) for the model behind this.
Admins can also create organization-wide API keys for agents that are not tied to a person, such as an internal bot. Those see only content shared with everyone. [Configuration](/advanced/configuration#api-keys) covers when to use which.
## Choose your setup
Claude on the web, Claude Desktop, and Cowork. One-click connector, sign in with your Hyperspell account.
ChatGPT on the web and desktop, as a custom connector.
Claude Code, Codex, Cursor, and any other agent that supports MCP.
Keep a Markdown copy of the brain on your machine for coding agents, and ask questions from the terminal.
**Slack.** If your organization uses Slack, an admin can also install the Hyperspell assistant in your workspace. Mention it in a channel or send it a direct message to ask the brain a question without leaving Slack. It appears as an option in Connect Agents once your Slack integration is connected. This is separate from the Slack integration that indexes your messages.
## Tips for getting good answers
* Ask the way you would ask a colleague. "Who is the point of contact at Acme?" works better than a keyword list.
* Mention names and dates when you have them. The brain resolves people, projects, and time ranges, so "what did Ben say about the migration in June" narrows things down a lot.
* Ask for sources when it matters. Answers cite the documents they came from, and you can open any of them in the dashboard.
* If an agent seems to ignore the brain, check that the connector is enabled for the current chat. Claude and ChatGPT both let you turn connectors on and off per conversation.