← 返回
未分类 中文

Openclaw mneme vision

Use local visual memory tools to index and search photos and videos from creator media libraries.
利用本地视觉记忆工具为创作者媒体库中的照片和视频建立索引并检索。
bkennedyshit bkennedyshit 来源
未分类 clawhub v1.0.0 1 版本 100000 Key: 无需
★ 0
Stars
📥 87
下载
💾 1
安装
1
版本
#latest

概述

Visual Memory

Use this skill when the user wants to index, search, organize, or inspect local

photo and video libraries.

When the user provides an uploaded photo/video/reel as an attachment, treat it

as temporary session context first. Use its local path to answer the current

question or run visual similarity search when appropriate. Do not call

media_index or otherwise add the attachment to durable visual memory unless

the user explicitly asks to add, save, index, catalog, or remember it.

Prefer media_index when the user points at a folder that has not been indexed.

Prefer media_search for natural-language searches such as finding shots,

finished videos, thumbnails, archive images, or clips matching a description.

Prefer media_search_by_image when the user provides an image and wants visually

similar items. Use media_describe when the user asks what is in a specific

media file.

Use gpu_status before GPU-heavy local tasks when the user is coordinating

video work, image generation, upscaling, VLM tagging, or another local model

process. Use gpu_release to unload resident Ollama models and hand VRAM to

that workflow, and use gpu_reclaim with the returned token when the workflow

is done. Use gpu_evacuate only when the user asks to free VRAM immediately

without creating a lease.

Return media results as previewable content cards when the host supports rich

output. Each result should include:

  • title: filename or human-readable clip name.
  • kind: image, video, audio, or unknown.
  • path: absolute local path, kept available for copy/reveal actions.
  • reason: short explanation for why this media matched.
  • timestamp or time_range: when available for video/audio hits.
  • preview: true when the file can be rendered by the host UI.

Mneme returns neutral media_artifacts.v1 JSON. Preserve the host application's

native styling; do not introduce custom colors, cards, gradients, or branding

unless the host already provides that UI.

Do not dump long bare paths as the main answer. Put the path behind a Copy path

or equivalent action when rich output is available, and show the user the actual

image/video/audio preview first. When rich output is not available, use concise

Markdown links or a compact table with path values.

Do not invent matches when the tool returns none.

Do not upload local media to a cloud service unless the user explicitly asks for

that separate action.

版本历史

共 1 个版本

  • v1.0.0 当前
    2026-06-07 06:48 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

knowledge-management

Baidu web search

ide-rea
使用百度AI搜索引擎(BDSE)进行网络搜索。适用于获取实时信息、文档资料或研究课题。
★ 246 📥 107,880
knowledge-management

Obsidian

steipete
操作 Obsidian 仓库(纯 Markdown 笔记)并通过 obsidian-cli 自动化。
★ 446 📥 105,191
knowledge-management

web-tools-guide

user_ec205dbb
MANDATORY before calling web_search, web_fetch, browser, or opencli. Contains required error-handling procedures (web_se
★ 73 📥 163,041