Benchmarking memory poisoning defense in AI agents
We ran MINJA-class injection attacks against Kalairos with trust scoring on and off. Here is what we learned about provenance, author-class tagging, and detection rates.
From the blog
Notes on building agent memory — benchmarks, design decisions, and tutorials from the team behind Kalairos.
Why versioned memory is non-negotiable for long-running agents
Stateless retrieval breaks the moment a fact changes. We walk through the Kalairos bitemporal model — event time, ingest time, valid intervals — and why time-travel queries matter for legal, medical, and finance agents.
Give your agent memory in three calls
init, remember, query. A 5-minute walkthrough of the Kalairos flat API, with a working example you can copy into your own agent today.