The Agent Memory Landscape

If you build with agents, "how should memory work?" stops being an academic question fast. This is a ready-made reading space for exactly that: 12 papers — from Generative Agents and MemGPT to HippoRAG and LongMemEval — each with a companion note, organized into themes with a suggested reading order, and an agent role that cites which paper a claim comes from.

1 · Clone it, fetch the papers

$git clone https://github.com/0-bingwu-0/stashbase-agent-memory
cd stashbase-agent-memory && ./fetch-papers.sh

The repo ships the notes; the script pulls the 12 PDFs straight from arXiv into papers/. Open the folder in StashBase and watch the space assemble itself: the PDFs gain extracted text in the background, and papers and notes become one searchable library — keyword search out of the box, and search by meaning once you add your own OpenAI or OpenRouter key.

2 · A map, not a pile

The papers are grouped into four themes — foundations (Generative Agents, Reflexion, Voyager, CoALA), memory architectures (MemGPT, MemoryBank, A-MEM, Mem0), retrieval as memory (HippoRAG and its sequel), and surveys & evaluation — with a reading order that starts from the map (survey + CoALA), moves through the classics and the systems, and ends by asking which benchmarks each system would actually pass. Every note keeps the same shape: what the paper proposes (grounded in its abstract), why it matters, and open questions to carry into the reading.

3 · Read with a companion that cites sources

The space's AGENTS.md gives the built-in agent a standing role: a research reading companion that searches the library before answering, names the paper (and note) behind every claim, and keeps a sharp line between what a paper says and its own inference. Ask "how do MemGPT and HippoRAG differ on retrieval?" and the answer comes back grounded in the actual PDFs sitting in your folder.

4 · Make it yours

The notes are companions, not substitutes — every one links the arXiv page. As you read, tell the agent what you think and it appends your take under a ## My notes section in that paper's file; when a new paper drops, it adds a note in the same format and a row in the README. The reading list stays alive because it's just files in a folder you own.