tadelstein9/two-tier-memory

Two-tier-memory is a small tool that gives AI coding agents a better way to remember things they have already solved, instead of relying on a growing pile of markdown files loaded into the agent's context at the start of every session. The README explains that this approach works fine with a handful of files but breaks down once there are a hundred or more, because everything competes for the same limited context space, and once that space runs out the agent starts quietly forgetting things and re-solving problems it already fixed. The fix here borrows an old idea from databases: instead of loading everything, keep two separate layers. The first layer is a small, always-loaded index file with just a title and pointer for each solved problem, so the agent knows what already exists without needing the full details. The second layer is a plain SQLite database where each hard problem is stored as a row, including the problem itself, its root cause, what worked, and any gotchas. The database can hold thousands of entries cheaply, since the agent only looks something up when it actually needs it, using a built in search feature. The tool is a single Python script with no external dependencies beyond Python's own built in SQLite support. It offers simple commands to set up the database, add a new solved problem, search past entries, look up a specific one, and regenerate the always-loaded index file from the database. The author notes that the tool alone does not solve the problem. It only works well if the agent is instructed to search the memory before trying to solve something again, and if entries are kept accurate and removed once they turn out to be wrong. This project is released under the MIT license.