Skip to content

[Proposal] Memory for ModelScope agents — 97.5% fewer tokens #942

Description

@vnbochkarev-netizen

Memory for ModelScope agents — proposal

Hi! ViBo is a modular memory layer for AI agents (semantic memory, web-search compression, living archive, private execution). It is model-agnostic and works with any Python agent.

Measured on 100K facts:

  • 50–150× less memory context (250,000 tokens → 263 per request)
  • 96.2% web-search token savings, 99.7% archive savings
  • Everything stays local; secrets (L1/L2/L3) never reach the LLM

We'd love to build an official integration for ModelScope agents: a vibo skill/tool that agents can call for persistent memory — vibo find before answering, vibo add after each session.

Available today: npm n8n-nodes-vibo, ClawHub @vnbochkarev-netizen/vibo-memory, GitHub: https://github.com/vnbochkarev-netizen/ViBo-memory

Would you be interested in a ModelScope integration guide or an official adapter? Happy to prepare it. Thanks!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions