xqli

That matches our experience pretty closely. A lot of “memory” issues we saw weren’t about storage capacity, but about what kind of information is allowed to persist and how it’s structured. Once everything is flattened into one blob, retrieval strategy becomes the only lever left — which is where vectors often get overused.

In Mneme, updates are intentionally asymmetric: – Facts are append-only and explicitly curated (they’re meant to be boring and stable). – Task state is rewritten as work progresses. – Context is disposable and aggressively compacted or dropped.

The idea is that only a small subset of information deserves long-term durability; everything else should be easy to overwrite or forget.

This reduces the need for heavy retrieval logic in the first place, since the model is usually operating over a much smaller, more explicit working set.

Mneme came out of long-running AI coding sessions where important state kept getting lost due to context compaction.

Instead of retrieval or embeddings, it treats memory as an explicit, structured artifact and separates: – stable facts – task state – ephemeral context

The goal is to make memory boring, inspectable, and durable across sessions.

Happy to answer questions or hear why this is a bad idea

Hi HN, I built Sbox because I needed a way to run untrusted / agent-generated code without adding policy logic, scoring, or heuristics.

Sbox is intentionally "zero intelligence": - no code analysis - no allow/deny rules - no LLM involvement

It only provides strong isolation primitives, and assumes all code is hostile.

I’m using it as a building block for long-running agents and automation systems. Would love feedback from people who’ve built sandboxes or secure runtimes.

Guys, any suggestions?

no, python is all my need^_^