nomarin-ui/agent-memory-ledger

Agent Memory Ledger is a Python tool that keeps a permanent, tamper evident record of everything an AI agent remembers, so that when an agent makes a bad decision based on old or wrong information, you can actually find out what it believed and why. Normal memory storage for AI agents just holds a current value, like a user's employer, with no record of when that fact was learned, whether it was ever double checked, or whether the agent even read it before acting on it. The ledger fixes this by recording every write and every read as a permanent entry. When you write a piece of information, you can attach where it came from. When the agent reads a piece of information later, that read is logged too, along with how old the information was at that exact moment. This lets you replay an incident afterward and see exactly what the agent believed right before it acted, and flag which of those beliefs were dangerously out of date. Nothing already written is ever changed or removed. If a fact is deleted, that deletion itself becomes a permanent entry, since the tool treats an agent forgetting something as a fact worth keeping. Every entry is linked to the one before it using a hash chain, so if anyone tries to edit or erase a past record afterward, a built in verification check will catch it. The system also tracks time in two ways: when something became true in the real world, and when the agent learned it, which lets you ask what the agent believed as of any past date with a single query instead of replaying history from scratch. You install it with pip and use either its command line tool or its Python API, which includes functions for writing, reading, checking history, verifying the record has not been tampered with, and resolving when two different names, like Acme Corp and ACME, refer to the same entity. The project has 81 passing tests, is not yet published to PyPI, and is released under the MIT license.