Persistent visual cache for LLM-driven software development. Caches screenshots using perceptual hashing, vector search, and AX trees to prevent token overhead and visual hallucination loops.
Latest release v1.4.2 · 8 Oct 2026
10 files for Claude Code, Codex, GitHub Copilot and Windsurf: state-memory-mcp, video-ingest, vision-memory-mcp, vision-memory-mcp copilot-instructions.md and 6 more — 45,077 tokens loaded in every session.
.github/copilot-instructions.md B 15,078 tok CLAUDE.md B 15,078 tok .windsurfrules B 14,921 tok .agents/skills/state-memory-mcp/SKILL.md A 45 tok .agents/skills/video-ingest/SKILL.md A 38 tok .agents/skills/vision-memory-mcp/SKILL.md B 34 tok .agents/skills/agent-reasoning-mcp/SKILL.md A 35 tok .agents/skills/behavior-mcp/SKILL.md A 28 tok .agents/skills/webcrypt-mcp/SKILL.md A 48 tok .agents/skills/world-model-mcp/SKILL.md A 39 tok These files are putervision/vision-memory-mcp's own configuration — they tell Claude Code, Codex, GitHub Copilot and Windsurf how to work on this repository, so they are not mods to install elsewhere. Copy one as a starting point and replace the parts that are about this project.