Four read, two write. The two that change access are the only ones with any
side effect. Everything else is read-only, enforced by a database role that
cannot write anything at all.
No model on our side
ResumeContext does not summarise your sessions or answer questions with a model of its own. The tools hand your agent the archive itself: passages, matches, transcript pages and query results, all verbatim. Your agent does the reading and the judging, the same way it would if every past conversation were a file on disk. Nothing it tells you was written by a second model it cannot see.Asking good questions
You do not need to name tools. These all work:- “What were we doing yesterday?”
- “Why did we drop the Redis cache?”
- “What’s the current staging database host?”
- “How many sessions has each person recorded this week?”
- “Read the end of my longest session.”
- “Invite priya@company.com to this project.”