nemilia.com

What It Is

Nemilia turns the browser into a full AI production environment: it runs as a single HTML file of about 3.5MB, with no account, no installation wizard, and no backend server. It can capture web content via a Chrome extension, while also supporting multi-agent workflows, document RAG, web search, MCP tool execution, and result export. Its positioning is closer to a “local-first AI workbench” than a simple chatbot.

Core Capabilities

Its biggest strength is being provider-agnostic: it can connect to cloud APIs such as OpenAI, Anthropic, Gemini, DeepSeek, Groq, and OpenRouter, and can also automatically detect Ollama, Jan, LM Studio, and Llamafile, with support for in-browser models such as WebLLM and Chrome AI. On the workflow side, it includes specialized agents such as SCOUT, LENS, QUILL, FORGE, and WEAVE, with support for parallel DAG execution, quality scoring, low-score retries, HITL human checkpoints, custom agents, Skills instruction sets, and an AI generator. For documents, it supports uploads of PDF, DOCX, TXT, CSV, MD, and other formats, combining BM25 with vector retrieval; for web research, it can integrate with Serper, Brave, Tavily, and SearXNG.

Pricing and Privacy

The page clearly states Free Personal Use, but does not disclose commercial licensing or enterprise pricing. Cloud models require users to provide their own API keys, so actual usage costs depend on the relevant providers. Privacy is one of its strong points: it has zero telemetry and runs client-side, with API keys encrypted locally using AES-256-GCM. Multi-profile workspaces also use AES-256-GCM and PBKDF2 with 200k iterations, and it states that in-memory data is cleared before switching profiles.

Pros and Cons

Its advantages are openness, control, and high feature density, making it especially suitable for turning local models, cloud models, document libraries, web capture, and MCP tools into production workflows. The downside is that configuration is relatively complex: users need to understand API keys, local models, MCP, RAG, and workflow design. Output quality also depends on the selected model, retrieval sources, and workflow configuration. Information on pricing, commercial support, and a Chinese interface is insufficient.

Who It’s For and Access from China

Nemilia is suitable for advanced knowledge workers in research, law, finance, market analysis, content production, code review, and similar fields, as well as developers who value local privacy and replaceable models. Access from China is not addressed in the main text, so it is considered unknown; payment methods are also not disclosed. If access or cloud model APIs are restricted, local Ollama/LM Studio can be considered, as well as alternatives such as Dify, AnythingLLM, Open WebUI, and Flowise.