What Prospektor reads, writes and never touches

For a security or AI policy review: what goes in, what comes out and what stays out of reach, as the product works today.

How to do it

To see every app it can reach, open Settings, then Connectors. Each stays off until your owner connects it. Gmail is the exception: each member decides for their own.
To see who can read your work, open Settings, then Team. Every member reads every pitch.
To write to your CRM, a member presses Push on one finished pitch, sees each change, then presses again. An assistant can push only with a key your owner made.
To remove your work, the owner uses Delete workspace. Nothing expires on its own.

It reads the public web, your site, and what your team types or uploads. Once connected: public Slack channels it joins, one read of a member's sent Gmail, access then revoked, and the CRM records a push checks, plus won deals or lists your owner imports.
It writes a company, up to ten people, a note and, where your CRM takes one, a deal; Slack gets a summary. It never sends email as you, and mails outsiders only an invite a member asks for.
Anthropic's Claude researches and writes. Prospektor trains no model on your work. Research on a public company is shared across workspaces and holds nothing of yours.
Staff screens show counts and dates. Engineers can read stored data during maintenance. The privacy policy on prospektor.ai names every service that receives data.