How is Folklore different from mem0?

mem0 is a single-user memory store that runs an LLM extraction on every write (a token/latency cost, and nondeterministic) and carries no signed provenance and no peer federation. Folklore writes with no per-write LLM call, ranks by signed provenance so poisoned context can be down-ranked, and federates so a team or network compounds. On a matched poison test, similarity-only ranking like mem0's is fully flipped (flip-ASR 0.625 toy, 1.0 at corpus scale); Folklore's provenance ranking drives it to 0.

Does Folklore need an LLM API key like mem0?

No. Folklore is CPU-only and needs no API key. mem0 needs an LLM (API key or a local model) because it runs an extraction on every memory write.

← Folklore · full comparison

Folklore: an open-source mem0 alternative

Q: Is there an open-source mem0 alternative?

Yes. Folklore is an open-source (MIT), local-first memory layer for AI agents and a direct alternative to mem0. It is CPU-only, needs no API key, and unlike mem0 it requires no LLM call per write, carries signed provenance, and federates peer-to-peer. Code and full benchmark: https://github.com/usefolklore/folklore.

Q: Does mem0 have signed provenance or federation?

No. mem0 is a single-user store ranked by embedding similarity with no cryptographic provenance and no cross-user federation. Folklore adds both.

Folklore is an open-source (MIT) alternative to mem0. The core difference: mem0 is single-user: memory never leaves your account. Folklore is peer-to-peer: agents share their research over the network, so a new agent starts from what the network already knows, not zero. It also writes with no per-write LLM call, is CPU-only with no API key, and ranks by signed provenance.

Star Folklore on GitHub →

Folklore vs mem0

Dimension	Folklore	mem0
License / cost	Open source (MIT), free	Open source core + hosted paid platform
Per-write cost	Local embed, no LLM call	LLM extraction on every write (tokens/latency)
Determinism	Deterministic writes	LLM-mediated, nondeterministic (recall varies run-to-run)
Signed provenance / poison-defense	Yes · flip-ASR → 0	No provenance field · flip-ASR 0.625 → 1.0 under poison
Peer federation	Yes (P2P, shipping)	No · single-user
Footprint	CPU-only, no API key, any MCP harness	Needs an LLM (key or local model)
Retrieval (BEIR SciFact NDCG@10)	0.7522, CPU · leads CPU retrievers	Parity at the retrieval layer (same embedder)

Figures from the public, reproducible benchmark; labeled measured vs simulated, negatives kept in.

When mem0 may fit better

mem0's LLM-mediated extraction can distill conversational memories into terse "facts," which suits chat-assistant personalization. If you specifically want LLM-summarized conversational memory and don't need provenance, CPU-only operation, or federation, mem0 is reasonable. Folklore is the better fit when you want poison-defensible, attributable memory that compounds across peers with no per-write LLM cost.

FAQ

Is there an open-source mem0 alternative?

Yes: Folklore (MIT, local-first, CPU-only). It adds what mem0 lacks: no per-write LLM call, signed provenance, and peer federation. Repo + benchmark.

Does mem0 have signed provenance or federation?

No: mem0 is a single-user store ranked by similarity, no cryptographic provenance, no cross-user federation. Folklore adds both.

Does Folklore need an LLM key like mem0?

No. Folklore is CPU-only, no API key. mem0 needs an LLM (key or local model) for its per-write extraction.

See the code + full benchmark →

Last updated: 2026-06-20. Folklore works alone (local) today; P2P federation is shipping. Every number traces to the public benchmark, negatives included. See the full memory-tool comparison.