Back to Gap Map Product / UX / Agent memory & context stores

Memori

Memori Labs
source available / Overall score: 2.8

Memori adds long-term memory to LLM applications by wrapping the model client, extracting facts, preferences, skills and agent-trace events from each conversation in the background and injecting the relevant ones into later prompts. Memories are kept in the developer's own SQL or document database or in Memori Cloud, linked by a graph of subject-predicate-object triples. Memori Labs builds it.

The repository was GibsonAI/memori before the project moved to Memori Labs, and memorisdk is the SDK's earlier PyPI name.

Openness

2 medium confidence
2.0
license
Apache-2.0 (Python and TypeScript SDKs and the Rust core: storage, embeddings and recall)+Proprietary (the Advanced Augmentation extraction service)
source
partial (the published core sends each conversation to Memori's hosted sdk/augmentation endpoint and writes back what it returns; the engine that extracts facts, preferences and triples is in no repository)
managed-tier
Memori Cloud (managed storage and dashboard; production plans sell managed, single-tenant or on-prem deployments)

Memori's SDKs and core, which store memories in the developer's database and recall them, are published under Apache 2.0. The step that decides what to remember is not: the SDK sends each conversation to Memori's hosted service to extract facts and relationships, metered by a free quota, so the memory logic can be neither read nor run locally.

Adoption

2 medium confidence
2.0

Adoption is the sum of monthly downloads of memori and memorisdk on PyPI and @memorilabs/memori on npm, the three packages that ship the SDK, most of it the memori package. Memori Cloud usage, the hosted MCP endpoint and the agent-host plugins are not counted.

Capability

4 medium confidence
4.0

Memori extracts typed memories from conversations and agent traces, keeps a knowledge graph beside them and merges repeated facts, all without changes to the calling code. Nothing in it retires a memory over time, and the extraction itself runs on Memori's hosted service.

Verified 2026-11-20, except some scores, still awaiting re-confirmation