Working personal system; continuing development.
A good AI session can still leave you starting over tomorrow. The assistant remembers part of the conversation, misses a decision, or confidently revives an idea you already rejected. I wanted the work to accumulate across projects and across assistants. Guru grew out of that frustration.
It is now part of my working environment: a place for decisions, research and corrections that should survive the conversation that produced them. Claude was my first AI collaborator and did substantial work with me. As I added other assistants, keeping useful knowledge independent of any one conversation became more important.
The uncomfortable discovery was that remembering more does not necessarily mean knowing more. A September 4, 2026 diagnostic found an accepted related item among the first ten results in 28 of 30 retrieval cases. That is a useful, limited result. It does not establish that the assistant answered the question correctly. The same audit exposed missing source links behind some stored facts, despite stronger claims in the documentation.
That changed what I ask of the system. I want to know where a statement came from, whether a later correction supersedes it, and when the evidence is too thin to use. Finding a familiar project name is only the beginning. The harder test is whether the next task benefits from the right decision, with its qualifications intact.
Guru is useful enough that I keep working with it, and imperfect enough that I keep questioning it. My next direction is a separate local librarian, inspired by Charles Sullivan’s work, to help examine and reconcile what has accumulated. That remains a research direction; it is not a demonstrated replacement for the assistants I use today. Read about the local librarian.
This personal Guru project is independent of and unaffiliated with the commercial Guru platform at getguru.com.