Open the HTML file. That's the entire setup. No npm, no Docker, no Python environment.
🤖
Multi-Agent Orchestration
Define specialist agents — researcher, analyst, writer, reviewer, synthesizer — and wire them into automated DAG pipelines. Each output is scored 1–10 and retried automatically if it falls short.
📄
Document RAG
Upload PDFs, DOCX, TXT, or CSV files. Nemilia chunks and embeds them with 384-dimensional vectors using hybrid semantic + BM25 search. Agents answer from your actual documents.
🌐
Live Web Research
NEXUS searches the internet in real-time via Brave, Serper, Tavily, or SearXNG. Findings are cited inline and passed to all downstream agents as grounded context.
👁
Human-in-the-Loop
Pause any pipeline at any agent. Approve the output, edit it inline, or send feedback to re-run that step. Loop as many times as needed before the pipeline continues.
🧠
Persistent Memory
Agents write MEMORY[key] = value in their outputs. These facts are stored and injected into all future agent prompts — building institutional knowledge across sessions.
📡
Offline AI via WebGPU
Run Llama, Phi, Qwen, Gemma, or Mistral entirely in your browser. No API key, no internet after model download. Models cache locally for instant future loads.
🔁
Drag-and-Drop Workflows
Build pipelines visually. Reorder agents by dragging, toggle HITL checkpoints per agent, and preview the execution flow as a live SVG DAG diagram before you run.
🎨
Visual Generation
VISION generates Chart.js charts, SVG diagrams, and HTML infographics — no image API required. Optional AI image generation via DALL·E, FLUX, or Stable Diffusion.
📁
Workspace Sync
Sync to a real folder on disk. Agents, prompts, and workflows become plain JSON and Markdown — editable in VS Code, version-controllable in Git, shareable with anyone.
