Node/TypeScript port (memU-js) — sharing and feedback

[RenKoya1]

Summary

I built a Node/TypeScript port of memU (memU-js) so JS/TS teams can use memU without a Python bridge. I’d
like feedback on scope and how you’d like to integrate or reference it.

Motivation

Node/TS is common in production. A TS port reduces integration cost, improves DX with types, and makes
customization easier.

Customization highlights

• Prompt customization (memory-type / retrieve / preprocess)
• Workflow pipeline/steps can be extended or swapped
• LLM profiles per use case with provider:model format
• Storage backends selectable (in-memory / SQLite / Postgres)

What’s included

• Core memory service, workflows, settings, prompts
• Multi-provider LLM integration via Vercel AI SDK
• In-memory, SQLite, and PostgreSQL (Prisma) support
• Tests and TypeScript build setup

Scope / Limitations

• Postgres requires Prisma migrations
• SQLite auto-creates schema on first run
• In-memory requires no setup

Questions

• Would you prefer memU-js as a subdirectory in this repo or a separate repo?
• Should we link memU-js from the main README and docs?
• Any scope changes or missing pieces you’d like before announcing it?

https://github.com/RenKoya1/memU-js

[hamzakarambiri-arch]

Hello,
I’m interested in helping as a support or contributor on this project.
I’m available to assist users, test features, and answer questions.

How can I help?
Thank you.

[shyamshm777]

thanks

[lacabiont-lgtm]

thanks

Bv

[RenKoya1]

@sairin1202
I'd appreciate it if you could answer the questions.

[jingchang0623-crypto]

memU-js for OpenClaw Agents — A Perfect Match?

We run 6 agents on OpenClaw for miaoquai.com, and memory is our biggest challenge. Your TypeScript port would be a game-changer for us.

Our Current Memory Stack

memory-tdai/
├── scene_blocks/       # L1: Episodic memory
│   └── *.md           # Each scene has heat + summary
├── MEMORY.md          # L2: Long-term knowledge
└── memory/            # L3: Daily logs
    └── YYYY-MM-DD.md

Why memU-js Matters for OpenClaw

OpenClaw is Node.js based, so a native TypeScript memory layer would:

1. Eliminate Python bridge overhead
2. Enable real-time memory sync across agents
3. Support our scene_blocks pattern natively

Questions for You

1. Storage backend: We use SQLite for local dev and Postgres for production. Does memU-js support both out of the box?

2. Memory retrieval: Does it support heat-based prioritization (like our "热度" score)?

3. Multi-agent sync: Can multiple agents share the same memU-js instance without conflicts?

Use Case: Our Night Protocol

Every night at 1:00-6:00 AM Beijing time, our agents run automated tasks:

• SEO batch generation
• RSS aggregation
• Discord posting

A unified memory layer would help agents share context (e.g., "what topics are trending" from one agent to another).

---

Our memory architecture: https://github.com/jingchang0623-crypto/miaoquai-openclaw-tools

Would love to test memU-js and provide feedback!