In daily development, Coding Agents generate massive amounts of conversations, tool calls, and code changes. But in reality, session data from different providers is often scattered across their own directories, with each format differing from the next.
To break the isolation between different agents' session storage, obelisk unifies these scattered records into a single queryable SQLite dataset. This not only lets humans observe through desktop applications, but also allows Agents to retrieve their own historical session content via SQL.
In this talk, we'll discuss some core design decisions behind building this indexing infrastructure. Once Agent historical data truly becomes structured and queryable, we can do quite a lot: failure pattern analysis, cross-session synthesis, and more.
Of course, since the foundation is SQLite, it presents no mental overhead for Python developers in the audience—if you want to analyze your session data, you just need to import sqlite3.