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

LongMemory

CaviraOSS
open source / Overall score: 2.6

LongMemory is a self-hosted memory engine that records what an agent learns as an immutable, time-stamped graph of entities, claims and relationships, then recalls it in strict, historical or associative modes within a token budget. One TypeScript engine runs as a library, CLI, HTTP and MCP server, dashboard and VS Code extension. CaviraOSS builds it, and it was previously published as OpenMemory.

Read against the repository's main branch, which holds the rewritten Hydrograph engine; the longmemory package on npm still ships the earlier OpenMemory code under the new name.

Openness

5 high confidence
5.0
license
Apache-2.0 (engine, CLI, server, dashboard, Python client and agent integrations)+MIT (n8n community node)
source
public
core features withheld
no — the 566-path tree at aec98ce carries the root Apache-2.0 LICENSE and per-integration Apache-2.0 or MIT files, with no ee/, enterprise/, commercial/ or proprietary/ path; the README offers no hosted edition or paid tier

LongMemory's engine, server, clients and integrations are published under permissive licenses, and the README presents the whole system as local-first and self-hosted, built from source or run from the published container. Nothing in the repository or its documentation describes a hosted edition or a paid feature held back from the code.

Adoption

1 low confidence
1.0

Adoption is read from npm downloads of longmemory, the package the README installs. The legacy openmemory-js package, the container image and the VS Code extension are not counted, and the published package carries the earlier OpenMemory engine rather than the current one.

Capability

5 medium confidence
5.0

LongMemory performs every part of the cycle itself, including retiring a fact once a later statement supersedes it and keeping the old one for historical questions. Its extraction is rule-based rather than model-driven, which keeps ingest deterministic but reads less from a conversation than the language-model pipelines of Memori or ReMe.

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