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

ReMe

AgentScope
open source / Overall score: 3.4

ReMe keeps an agent's long-term memory as ordinary Markdown files, distilling conversations and resources into daily notes and consolidating them on a schedule into a linked knowledge base of facts, preferences and procedures. Agents reach it through a Python API, CLI, HTTP, MCP or plugins for hosts such as QwenPaw, Hermes Agent and OpenClaw. The AgentScope team builds it; it grew out of MemoryScope.

Openness

5 high confidence
5.0
license
Apache-2.0 (ReMe, ReMe Studio, integrations and plugins)+MIT (two bundled benchmark judge plugins)
source
public
core features withheld
no — the 1,592-path tree at 3db84c7 carries the root Apache-2.0 LICENSE, Apache-2.0 files for Studio and plugins and MIT files for the memlens and mobilemem judges, with no ee/, enterprise/, commercial/ or proprietary/ path

ReMe and its Studio, integrations and plugins are published under Apache 2.0, apart from two MIT-licensed benchmark judges, and the complete system runs locally from pip, source or Docker against the user's own model keys. No hosted edition or paid tier is offered beside it.

Adoption

3 medium confidence
3.0

Adoption is read from monthly PyPI downloads of reme-ai, the package every install path uses. Downloads pulled in by host plugins are not separated from direct installs, and the retired MemoryScope package is left out.

Capability

4 medium confidence
4.0

ReMe distills conversations into daily notes and later folds them into long-term pages, correcting a page when new material contradicts it. Memories never expire or fade by age; only raw resource folders can be given a retention period.

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