Remem — Memory for AI Engineers

Re (in Act) + Mem (ory). Persistent memory layer for AI Engineers built on Claude Code / OpenCode and the Re in Act spec.

Zero databases. All memory is plain Markdown files in the .agents/ standard directory.

Quick Start

git clone https://github.com/RIA-Spec/remem.git
cd remem
# See remem-plan.md for full implementation details

Structure

.agents/memory/MEMORY.md   — Project facts (always injected, 5K chars / 200 lines)
.agents/memory/TEAM.md     — Organization facts (3K chars / 100 lines)
.agents/memory/USER.md     — Current user preferences (2K chars / 80 lines)
.agents/memory/People/     — Per-person knowledge (lazy loaded)
.agents/memory/Sessions/   — Episodic memory (on-demand search, uncapped)
.agents/skills/            — Procedural memory via standard SKILL.md

Prompts

• prompts/init-memory.md — Bootstrap memory from enterprise sources
• prompts/extract-facts.md — Extract facts from ACP trajectory
• prompts/dream.md — Background consolidation (expiry, merge, trim)

Tech Stack

• TypeScript + @mcpc-tech/acp-ai-provider
• Claude Code / OpenCode (via ACP)
• reason CLI (Re in Act)
• ripgrep (or grep)

Spec

Memory system design follows the Re in Act open specification — see remem-plan.md.