Interactive visualization and exploration of large language model (LLM) systems, infrastructure, and inference workflows
-
Updated
Sep 13, 2026 - JavaScript
Interactive visualization and exploration of large language model (LLM) systems, infrastructure, and inference workflows
Plataforma NEXUS para selección de contexto, enseñanza sobre cambios de código y memoria durable local-first.
Bio-inspired semantic filtering engine. Ecosystem simulation for knowledge quality assessment.
engramx's bi-temporal mistakes + structural code-graph as auto-discoverable Claude Code Skills. Catches recurring mistakes before agents repeat them. Apache 2.0, local SQLite, zero cloud.
Unofficial community plugin: per-session Engram memory for DeepSeek Harness — spawns an engram MCP child per agent session rooted at the session workspace, registers mem_* tools per agent scope, and injects the Memory Protocol as a system-prompt section.
To associate your repository with the engram topic, visit your repo's landing page and select "manage topics."