The agent

The Knowledge page (the Chat section in the sidebar) is home to Hypr, Hyprcore's reasoning agent. Unlike session chat — which answers questions about a single meeting and only reads — the agent works across your entire vault and can take actions on your behalf.

Ask it something, and it works in steps: it thinks, calls tools to retrieve or change things, reads the results, and repeats until it has an answer. You see the whole trace as it happens — each thought, each tool call, each result.

What it can do

Read across everything. The agent searches and reads your recorded meetings, transcripts, people, action items, and captures to answer questions grounded in what was actually said — no guessing from memory. It can also look at your calendar for upcoming meetings and pull open commitments (promises and action items with owners and due dates) from across every meeting.

Create and edit notes. The agent can create a new note in your vault or replace the body of an existing one (and rename it). Handy for turning a request like "draft a follow-up from today's call and save it to my Inbox" into a real note on disk.

Run a sandboxed command. The agent can run a shell command on your Mac inside a kernel sandbox — writes are jailed to a scratch workspace and credential files are unreadable — for read-only diagnostics or computing over files.

Remember you. The agent keeps a small persistent memory across conversations: durable facts like your preferences and the key people and projects you work with. It's for stable context, not task progress.

Approval before it acts

Reading is frictionless — the agent retrieves whatever it needs to answer you. But anything that changes your machine or your vault pauses for your explicit approval first. Three tools are gated:

• Create note

• Update note

• Run command

When the agent wants to use one, the run stops and an approval card appears showing the tool name and the exact arguments (the note it's about to write, the command it's about to run). You Approve or Deny. Nothing happens until you decide, and if you leave the card sitting the request is treated as denied rather than hanging. You can also Stop a run at any point.

Memory is intentionally not gated — saving a preference is low-stakes, so it happens without a prompt.

Note: The agent's conversations and its full reasoning trace are saved in your vault, so reopening a chat brings back everything it did and why.

How it differs from session chat

	Session chat	The agent
Scope	One meeting's transcript and notes	Your whole vault, plus calendar
Actions	Read-only — answers questions	Reads and acts (notes, sandboxed commands, memory)
Where	The Chat button on a meeting page	The Knowledge page (Chat in the sidebar)

Reach for session chat when your question is about a specific recording. Reach for the agent when you want an answer that spans many meetings, or when you want Hyprcore to actually do something with what it finds.