← Back to Skills Marketplace
vemem — visual entity memory
by
linville-charlie
· GitHub ↗
· v0.1.1
· MIT-0
71
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install vemem
Description
Visual entity memory — remember faces, objects, and places across sessions with persistent identity. Use when the user asks who is in an image, when you need...
Usage Guidance
This skill appears coherent and transparent for face/object identity storage. Important things to consider before installing: 1) It stores biometric vectors locally (default ~/.vemem) and will download ~200MB InsightFace weights on first image use — run in a dedicated venv and use a test VEMEM_HOME to inspect behavior. 2) Do not enable the OpenClaw sidecar integration unless you want an always-on image-processing layer; that sidecar can auto-process every image your agent sees. 3) The MCP tool accepts image_path inputs — avoid passing untrusted filesystem paths to prevent unintended local file reads. 4) If you plan to compose with remote LLM/VLM services (OpenAI/Anthropic), be aware that images and derived text may be sent to those services and that you are responsible for consent and regulatory compliance (GDPR/BIPA/CCPA). 5) Pin package versions, audit the pip-installed files (pip show -f vemem), and confirm the GitHub repo and release signatures before using in production.
Capability Analysis
Type: OpenClaw Skill
Name: vemem
Version: 0.1.1
The 'vemem' skill provides visual entity memory and face recognition, which involves the collection and storage of sensitive biometric data. While the documentation is transparent and includes GDPR compliance details, the skill instructions require the agent to perform high-risk actions: installing an external Python package, accessing the local filesystem via 'image_path' parameters in SKILL.md and mcp-tools.md, and performing a network download of model weights. These capabilities are plausibly necessary for the stated purpose but constitute a significant attack surface and privacy risk, fitting the criteria for a suspicious classification.
Capability Assessment
Purpose & Capability
Name/description (visual entity memory) match the SKILL.md and example code: the skill instructs use of a Python package that encodes images to embeddings, stores them in a local LanceDB store (~~/.vemem) and provides recall/identify/forget primitives. No unrelated binaries, env vars, or credentials are requested.
Instruction Scope
SKILL.md is explicit and scoped: it only processes images you pass to it and documents local paths (~/.vemem, ~/.insightface). Examples show optional composition with remote LLM/VLM APIs (OpenAI/Anthropic) which would send image data if you choose those providers; the skill itself does not require API keys. One operational caution: the MCP server API accepts an image_path argument (local filesystem path) — if you call that with arbitrary paths, the server will read them, so agent callers should avoid passing untrusted paths. The sidecar/OpenClaw integration can auto-process every image if explicitly installed and registered — the README calls this out as a separate opt-in.
Install Mechanism
The registry package is instruction-only (no install spec), but runtime usage requires 'pip install vemem' and a ~200MB InsightFace weight download on first use. Both actions are typical for a vision library; the SKILL.md points to the project's GitHub and recommends pinning versions and venv usage. Network activity is limited and documented (first-use model download).
Credentials
The skill declares no required env vars or credentials. The SKILL.md optionally references VEMEM_HOME and VEMEM_MCP_ACTOR for configuration; example recipes show how separate third-party LLM/VLM API keys would be used if the deployer composes them, but those are not required by the skill itself.
Persistence & Privilege
always is false and the skill does not request forced permanent inclusion. The SKILL.md documents long-lived local storage (~/.vemem) containing biometric embeddings and provides forget/restrict/export primitives; the sidecar that can be always-on is explicitly called out as separate opt-in behavior.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install vemem - After installation, invoke the skill by name or use
/vemem - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.1
vemem 0.1.1
Added up-front privacy/scope section to address the automated review's transparency concerns.
v0.1.0
Initial release of vemem: persistent visual entity memory.
- Bridges vision models and text LLMs, turning images into stable, named entity references with attached context.
- Supports face/object detection, identity assignment, fact storage per entity, and persistent recognition across sessions.
- Provides robust API for observing, labeling, identifying, recalling, correcting, merging, splitting, and forgetting visual entities.
- Ensures privacy compliance with GDPR/BIPA (supports hard delete and export for user data).
- Designed for easy integration via MCP server or direct Python use.
Metadata
Frequently Asked Questions
What is vemem — visual entity memory?
Visual entity memory — remember faces, objects, and places across sessions with persistent identity. Use when the user asks who is in an image, when you need... It is an AI Agent Skill for Claude Code / OpenClaw, with 71 downloads so far.
How do I install vemem — visual entity memory?
Run "/install vemem" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is vemem — visual entity memory free?
Yes, vemem — visual entity memory is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does vemem — visual entity memory support?
vemem — visual entity memory is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created vemem — visual entity memory?
It is built and maintained by linville-charlie (@linville-charlie); the current version is v0.1.1.
More Skills