Repository navigation
AI-Memory-Book: local, cited conversation recall for Codex over MCP #61668
Afloat16
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
I maintain AI-Memory-Book, an MIT-licensed Python tool for keeping a local, searchable archive of coding conversations.
The workflow I want to support is continuing a project in a fresh Codex session and recovering an earlier constraint, a corrected parameter, or an approach that already failed. Conversations are organized by project and task in SQLite. The MCP interface returns bounded evidence with source references, so the agent can read the relevant original text before continuing.
What it provides
A small engineering example
The repository includes synthetic sessions covering a data-input contract, a failed shortest-path approach, and a KV cache parameter corrected from 128 to 32. Another project has a different value for the same parameter, so the example also exercises project scoping.
After installing from the current
mainsource following the README, run this from the repository directory:memory-book --db ./demo-memory.db import examples/engineering-workflow.json --format canonical memory-book --db ./demo-memory.db recall "current kv_cache block_size" --project inference-demoFor Codex, the MCP setup guide explains how to connect the local server. The engineering workflow guide walks through the examples and source lookup.
This is an early project. The published retrieval checks use development datasets and should not be read as evidence of general coding-productivity gains. I would especially appreciate feedback on the Codex setup and concrete cases where an earlier decision is still retrieved incompletely or an outdated value is returned.
Repository: https://github.com/Afloat16/AI-Memory-Book
All reactions