MemMachine (paper)
https://arxiv.org/abs/2604.04853
Stores raw conversational episodes without LLM-driven extraction, indexing at sentence level; retrieval expands nucleus matches into neighboring-context clusters to address the embedding-dissimilarity problem in dialogue data. A Retrieval Agent routes queries across direct, parallel-decomposition, or iterative chain-of-query strategies. Distinct from the MemMachine framework already listed above (Dedicated memory layers).
Active · Verified 2026-10-28 · fresh as of 2026-10-28 · 21 cited facts, 0 with a third-party source · Suggest a correction
At a glance
- Tier
- T4
- Latest release
- not applicable — not OSS fix
- License
- not applicable — not OSS fix
- GitHub
- not applicable — no GitHub repo fix
- Pricing
- not applicable — not commercial fix
- Funding
- not applicable — not commercial fix
- Freshness
- 2026-10-28 fix
- Layer
- agent-infra repo fix
Memory taxonomy
- storage
- vector
- retrieval
- similarity
- persistence
- long-term
- update
- append-only
- unit
- episode
- governance
- inspectable
- conflict
- append
When to use
Optimised for: not applicable - research paper
Anti-fit: not applicable - research paper
Pros & cons
Pros
Ground-truth preservation avoids the factual hallucination risk of extraction-based condensation; cluster-expansion retrieval directly targets the embedding-dissimilarity gap in conversation data.
Cons
Storing raw episodes grows storage linearly with conversation length; no published evaluation on months-scale interactions.
Claims & capabilities
0.9169 on LoCoMo (gpt-4.1-mini) — above Mem0, Zep, Memobase, LangMem. 93.0% on LongMemEvalS. ~80% fewer input tokens than Mem0 in matched comparisons.
Technical surface
- API surface
- not applicable — research paper
- Backend storage
- not applicable — research paper
- Deployment
- not applicable — not a deployable product
- Embedding model
- not applicable — research paper
- Multi-tenancy
- not applicable — research paper
- MCP
- not applicable — research paper, no deployed product
- A2A
- not applicable — research paper, no deployed product
- OpenTelemetry
- not applicable — research paper, no deployed product
Similar systems
Other recent method papers — theorized, no distinct product in the catalog, ranked by inbound references.
- Compressive Transformer T3
Maintains recent states in full resolution while compressing older memories with learned compression functions. DeepMind.
- MemGPT v2 / agent-tools T3
Already in catalog as the foundational MemGPT paper. Note: Letta is the productionised successor (cross-listed).
- Transformer-XL T3
Extends context through segment-level recurrence + caching of hidden states from prior segments. Foundational long-context architecture.
- Generative Agents T3
Park et al. — landmark agent-simulation paper. Reflection + memory stream + retrieval enable believable agent behavior.
- MemoryBank T3
Enhances LLMs with long-term memory. Early influential paper.
- Reflexion T3
Language agents with verbal reinforcement learning.
Sources (20 cited facts, 0 third-party)
Each fact links to the page it came from; the badge says who published it (how to read sources).
- What it is Ground-truth-preserving episodic + cluster-expansion retrieval vendor
- What it does & what's distinct Stores raw conversational episodes without LLM-driven extraction, indexing at sentence l… vendor
- Claims / benefits 0.9169 on LoCoMo (gpt-4.1-mini) — above Mem0, Zep, Memobase, LangMem. 93.0% on LongMemEv… vendor
- Modalities text vendor
- Created 2026-04 vendor
- Citations 2 cites 2/yr paper
- Founders / pedigree no data — searched arxiv.org/abs/2604.04853; author affiliations not found in preview vendor
- Memory volume / scale no scale claim — research vendor
- Performance 0.9169 on LoCoMo (gpt-4.1-mini) — above Mem0, Zep, Memobase, LangMem. 93.0% on LongMemEv… vendor
- Reproducibility Too recent vendor
- Versioning immutable append-log vendor
- Tombstoning append-only no delete vendor
- Schema evolution schema with migration vendor
- Contradiction handling always append (no resolution) vendor
- Forgetting policy none (permanent) vendor
- Pros Ground-truth preservation avoids the factual hallucination risk of extraction-based cond… vendor
- Cons Storing raw episodes grows storage linearly with conversation length; no published evalu… vendor
- Project & sources arxiv 2604.04853 vendor
- citation-trajectory [] paper
- taxonomy-three-layer-rollup agent-infra repo
Related systems
References (1)
- LoCoMo cites — S2 isInfluential citation