Zero-LLM memory
Ingestion and retrieval are deterministic: no extraction, no summarization, no model dependency.
Memory you can unit test.
Zero LLM calls.
CoreMem is a deterministic memory core for agents that need to remember without rewriting user history through a model. Store raw memories, retrieve them with transparent heuristics, and build wake-up context you can inspect.
Agent memory that stays explainable.
Wake-up context stack
Identity, essential memories, session context, and deep search form a layered memory surface.
Heuristic re-ranking
Keyword overlap, recency, temporal cues, names, and quoted phrases improve result ordering.
Backend choice
Use pure ChromaDB for a light baseline or HybridDB for keyword plus semantic retrieval.
Batch ingestion
Load large message histories efficiently for benchmarking and real user memory import.
Benchmark-ready
Includes LongMemEval evaluation paths for reproducible memory recall testing.
Architecture
A memory core, a backend, and deterministic context builders.
CoreMem separates the public memory API from storage backends and retrieval heuristics. The agent asks for wake-up context or deep search; CoreMem decides what to fetch and how to rank it.
MemoryCore
The main API for ingesting memories, searching, counting, clearing, and building context.
StoreBackend
A shared interface implemented by ChromaBackend and HybridBackend.
SearchHeuristics
Deterministic post-retrieval scoring for keyword, temporal, recency, name, and quote signals.
WakeUpContext
Composes L0 identity, L1 essentials, L2 session context, and L3 deep search.
The value prop: memory that behaves like software, not magic.
check_circlePreserves raw text verbatim rather than rewriting memories through a model.
check_circleDesigned for agents that need persistent user context across sessions.
check_circleCan start simple with ChromaDB and move to HybridDB without changing the API.
check_circleBest positioned as an inspectable memory core, not a black-box personalization platform.
Compared to alternatives
A memory layer for developers who care why something was remembered.
Many memory products optimize for automatic personalization. CoreMem optimizes for determinism, cost control, local privacy, and testability.
Versus mem0 and Zep
Those platforms provide polished managed memory and graph/context features. CoreMem is smaller and more inspectable: raw records, deterministic scoring, and no hidden inference call during retrieval.
Versus Letta/MemGPT
Letta is a broader stateful-agent runtime. CoreMem is a lower-level memory substrate that can sit under any runtime when you want explicit retrieval behavior.
Versus LangGraph checkpoints
Checkpointing saves execution state. CoreMem retrieves long-term episodic context. They complement each other rather than competing directly.
Versus hosted RAG memory
Hosted file search is convenient but provider-bound and opaque. CoreMem keeps memory local, framework-neutral, and cheap to run repeatedly.
Use it when
Use CoreMem when repeatability beats magic.
radio_button_checkedUse it for local agents that need durable user context without LLM-based extraction or summarization.
radio_button_checkedDo not claim proven benchmark leadership until committed benchmark result artifacts exist.
radio_button_checkedPosition it as an experimental, inspectable memory core rather than a finished managed memory cloud.
radio_button_checkedThe clearest CTA is not 'make your agent smarter automatically'; it is 'make memory predictable enough to test.'
Add memory without adding hidden inference.
Use CoreMem when every remembered result should be explainable, repeatable, and cheap enough to call often.
Add deterministic memory