memory system for 24/7 proactive agents.
what it is
memU treats memory as a persistent, versioned, searchable knowledge graph instead of ephemeral context. built for agents that run continuously across channels (slack, discord, telegram, SMS) and need to remember things across sessions, not just within a conversation.
the problem it solves
every personal AI project hits the same wall:
- you tell your agent something once
- it forgets
- you tell it again
- it remembers but pulls it into the wrong context
- you manually edit MEMORY.md
- it ignores the edit
memory isn’t a token window problem. it’s an architecture problem.
how it works
instead of “conversation history” (append-only log of turns), memU provides:
- queryable, prunable knowledge graph of facts, preferences, and decisions
- version control for memory edits
- semantic search across sessions
- context-aware retrieval (knows when to surface what)
when your agent runs 24/7, memory can’t be “context window management.” it has to be a database.
use cases
- openclaw / moltbot / clawdbot deployments
- multi-channel agents that need consistent memory across platforms
- long-running research agents that accumulate knowledge over weeks/months
status
active development, production-ready. targets openclaw and compatible frameworks.
the shift from chatbots (conversation history) to agents (memory systems).