GitHub Repository

Browser-native AI workspace — multi-agent orchestration, HITL review, semantic vector RAG, drag-and-drop workflows. Single HTML file, zero backend.

9 starsHTML

Nemilia – multi-agent AI workspace in a single HTML file, no back end

Name: Nemilia – multi-agent AI workspace in a single HTML file, no back end
Availability: InStock
Author: Nemilia

by Nemilia·Mar 5, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●●BangerZero to OneWizardryBig Brain

Single-file multi-agent workspace with MCP tools, WebGPU offline mode, full data sovereignty.

Strengths

•Genuine architectural innovation: entire agent runtime + orchestration + RAG in one HTML file eliminates deployment friction.
•MCP protocol integration means agents can execute real tools (filesystem, APIs, code) without server infrastructure.
•WebGPU offline inference + hybrid RAG (vector + BM25) shows thoughtful technical layering for a constrained runtime.

Weaknesses

•Browser vendor lock-in for features like WebGPU limits long-term portability and cross-platform maturity.
•No obvious path to team collaboration or workspace sharing without building sync/persistence layer.

Post Description

A full multi-agent AI workspace in a single HTML file. No backend. No install. No build step. No account. You own the entire runtime — open it in a browser and it works.

That's not a constraint. That's the point.

Nemilia is built around the idea of AI sovereignty — your agents, your workflows, your data, your keys, all running on your machine. Nothing phoned home, nothing locked to a platform, nothing that breaks when a SaaS company changes its pricing.

What you get: - Build custom agents with their own roles, personalities, system prompts, and model overrides - Design custom workflows with drag-and-drop pipeline builder — chain agents in any order, set HITL review checkpoints, run parallel DAG execution - MCP tool execution — build agents that call external tools in real time (filesystem, databases, APIs, code execution) over streamableHttp or SSE - Document RAG with hybrid Transformers.js vector embeddings + BM25 keyword search, fully client-side - WebGPU offline mode — run Llama, Qwen, Phi, Gemma entirely in the browser, no API key, no internet - Connect any provider: OpenAI, Anthropic, Groq, Gemini, Mistral, DeepSeek, Ollama, LM Studio — swap anytime, your workflows follow - API keys encrypted AES-256-GCM, never leave your browser - Workspace syncs to a real folder on disk as plain JSON and Markdown — version control it, edit it in VS Code, take it anywhere

One file. Any browser. Your AI team.

Getting started takes about 60 seconds: download the file, open it, paste any API key, point it at LM Studio/JAN, or load up a model directly into your browser via WebGPU.

Released it publicly last week. Would genuinely love feedback.

Site: https://nemilia.com GitHub: https://github.com/luislopez1212/nemilia

P.S. Nemilia means "to think, to remember, to imagine" in Nahuatl, the language of the Aztecs. Felt like the right name for something that does all three.