Capwave is not a chatbot, and this page is not against them: I run on the same class of models. ChatGPT, Claude, Copilot and the newer personal agents such as Instinct are single-player by design. They learn one person, inside one conversation, from what that person pastes in. I am built the other way round: one memory for the whole firm, connected to the inbox, the calendar and the notetaker, and it stays when people leave.
Brilliant on demand. Ask it anything and it answers, once, for you, from what you gave it. Nothing arrives on its own, and nothing is remembered by the firm.
Working before you ask. I read what reached the firm overnight, score it, brief the meeting, draft the follow-up and hold it until it closes, for every seat at once.
Keep the assistant for the one-off. Ask me from inside Claude or ChatGPT (MCP) when you want the firm's memory in the chat you already use.
Assistant descriptions are general and refer to the products' publicly described plans as of September 2026. ChatGPT, Claude, Copilot and Instinct are products of their respective companies; Capwave is not affiliated with them.
This is the part a chat cannot do: the brief that was waiting before the meeting, built from the deck, the data room, the last call and what you promised in March.
"The USV investment process looks like an ongoing team conversation, and we now have agents that are part of that conversation."
"So much of what a venture capitalist firm is are its relationships with the world, because to build a company, you end up needing a lot of relationships."
Public statements, quoted with attribution. Not endorsements.
Because it knows one person. Each partner's assistant learns that partner, and the firm is the sum of what they never told each other. I am one memory for the whole firm: every deal, score, promise and relationship, readable by every seat, and it stays when people leave. A personal assistant makes one person faster. I make the firm know what it knows.
Yes. Connect me through MCP and ask about any deal, person or promise from the chat you already use. The answer comes with sources.
Yes. I run on frontier models and own none of them. What is mine is the memory, the connections to your inbox, calendar and notetaker, the scoring against your thesis, and the follow-through. Your data trains nothing.