Search Results for "muse"

MUSE gives AI coding agents persistent cross-session memory and multi-role governance through plain Markdown files. Supports Claude Code, OpenClaw, Cursor, Windsurf, Gemini CLI, and Codex via one-command install. Built-in MCP Server for programmatic access. 56 skills, auto memory capture, semantic compression, role-based governance, multi-project management.

MUSE is a framework for learning multilingual word embeddings that live in a shared space, enabling bilingual lexicon induction, cross-lingual retrieval, and zero-shot transfer. It supports both supervised alignment with seed dictionaries and unsupervised alignment that starts without parallel data by using adversarial initialization followed by Procrustes refinement.

Muse Glimmer-30B is Meta Superintelligence Lab’s open-weight multimodal model built specifically for autonomous agentic tasks on consumer hardware. Distilled from the larger Muse Spark, it combines multi-step reasoning, reliable tool use, coding, failure recovery, and image understanding in a dense 29.6B-parameter architecture with a dedicated 1.8B-parameter perception encoder.