MemoryLake vs Dume
Dume offers a lightweight AI memory service for developers prototyping personalization. MemoryLake is the better fit when memory must persist across models, agents, sessions, and enterprise systems with governance, provenance, and verified accuracy.
Dume
Lightweight AI Memory Service
Strengths
- Minimal API surface — fast to prototype memory features
- Simple pricing suited for early-stage experimentation
- Good for single-app personalization use cases
- Low barrier for developers testing memory ideas
- Friendly developer experience and docs
Limitations
- No structured memory types for precise reasoning
- No Git-like versioning, branching, or rollback
- Provenance and enterprise governance are limited
- Not benchmarked on long-term memory evaluations
- Memory is typically bound to the app that writes it
MemoryLake
AI Memory Infrastructure
Strengths
- 6 structured memory types designed for AI reasoning
- Git-like memory versioning with safe rollback
- Source-level provenance on every memory
- 94.03% accuracy on LoCoMo long-term memory benchmark
- Multimodal ingestion from documents, databases, APIs, and media
- Enterprise controls: SOC 2, ISO 27001, GDPR, CCPA
Considerations
- Richer abstractions than a minimal memory API
- Greatest value when memory must serve multiple products and models
- Pricing depends on deployment shape and workload
Feature-by-Feature Comparison
| Feature | Dume | MemoryLake |
|---|---|---|
| Primary purpose | Lightweight memory API | AI memory infrastructure |
| Memory model | Untyped user facts | 6 typed memory categories |
| Versioning | Not a first-class feature | Git-like history, branching, rollback |
| Provenance | Basic metadata | Source-level provenance per record |
| Multimodal ingestion | Primarily text | Text, docs, tables, images, audio, video, DBs, APIs |
| Conflict handling | Manual or overwrite | Automatic conflict detection + resolution |
| Accuracy (LoCoMo) | Not published | 94.03% overall on LoCoMo |
| Enterprise controls | Limited | SOC 2, ISO 27001, GDPR, CCPA |
| Portability | Tied to the Dume API surface | Portable across any LLM and agent framework |
| Best fit | Prototyping personalization quickly | Durable AI memory across products and agents |
Architecture Comparison
Dume focuses on a simple API for storing and retrieving user facts. MemoryLake is a full memory lake with structured types, provenance, versioning, and multi-source ingestion.
Dume Pipeline
MemoryLake Pipeline
Which Is Right for You?
Choose Dume if...
- You are prototyping personalization in a single app
- You want the smallest possible memory API surface
- You do not yet need versioning, provenance, or governance
- Memory portability across models is not a priority
- You prefer to compose memory logic in your own code
Choose MemoryLake if...
- You need a real memory system of record, not just a key-value API
- You require versioning, provenance, and audit trails
- You ingest from documents, databases, APIs, and multimedia
- You need benchmark-verified accuracy (94.03% on LoCoMo)
- You need enterprise compliance for serious production use
- You want memory that travels with the user across models and products
Frequently Asked Questions
Is Dume a good product?
Yes, for early-stage developers who want a minimal memory API for personalization prototypes.
Who should choose Dume?
Teams prototyping personalization in a single app who want the simplest possible memory surface.
Who should choose MemoryLake?
Teams building AI products where memory must persist and be governed across models, agents, and sources.
Can we migrate from Dume to MemoryLake?
Yes. The typical trigger is outgrowing a minimal memory API and needing types, versioning, and cross-model portability.
Is MemoryLake open-source?
No, it is a managed platform with API access.
Does MemoryLake replace RAG?
No. It works alongside RAG as a durable memory layer.
Does MemoryLake publish benchmarks?
Yes — 94.03% overall on LoCoMo with segmented scores.
Is MemoryLake only for enterprise?
No. It shines there, but it fits any team building multi-agent or multi-product AI.
What about pricing?
Dume offers simple developer pricing. MemoryLake pricing depends on deployment shape.
Biggest difference?
Dume is a minimal memory API. MemoryLake is a memory system of record for AI.
Ready to Try MemoryLake?
Move from a minimal memory API to a durable memory system of record — 6 typed categories, Git-like versioning, source-level provenance, and 94.03% LoCoMo accuracy.