maxgfr

Aparte — AI skill for Claude Code

AI community

Private AI chat that runs entirely in your browser: your own keys or on-device models (WebLLM, Ollama), MCP tools, skills, Python sandbox, local RAG, voice.

How to install Aparte

This entry records only its repository, not the path inside it, so there is no exact command to give. Open maxgfr/aparte and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Aparte does

Private AI chat that runs entirely in your browser: your own keys or on-device models (WebLLM, Ollama), MCP tools, skills, Python sandbox, local RAG, voice. Encrypted at rest and synced end-to-end to your own Git repo. PWA + Rust CLI.

Alternatives in AI

Guaardvark — The self-hosted AI studio: local video, image, music, voice, LoRA training, coding swarms, RAG and screen agen 211 ★
Friday Showcase — Friday — A 24/7 AI Assistant Built Entirely on Claude Code 181 ★
Mnem — Git for AI Agent Knowledge 144 ★

README

Aparté

Aparté is a private AI chat that runs entirely in the browser. It talks to the models you choose, with your own API keys or on your own hardware, and keeps conversations, documents and photos encrypted on the device, synced end to end to storage you own. There is no account and no server: the app is a static site, and a Rust CLI opens the same vault from the terminal.

**App:** (installable as a PWA on desktop, Android and iPhone; the shell works offline).

Features

Models: Anthropic (or an Anthropic-compatible base URL, with or without its /v1), OpenAI, Google Gemini, Mistral, OpenRouter and over a hundred OpenAI-compatible services of the llm-models catalog with your own keys; an option lists the providers a browser cannot call directly, for use through a proxy you run. Ollama, LM Studio and Jan on your computer; models in the browser with WebLLM; Chrome's built-in Gemini Nano. Streaming, reasoning display, effort levels, regenerate and edit as branches.
Tools: built-in web reading, MCP servers (Streamable HTTP, OAuth 2.1 with PKCE) with approvals that show the full arguments and detect definition changes, skills (SKILL.md), and Deep research with keyless search backends and checked citations.
Sandboxes: JavaScript and Python (Pyodide with numpy, pandas and matplotlib) in sandboxed frames, and HTML, SVG and Mermaid artifacts that run only after a click.
Knowledge: local knowledge bases over your documents (PDF and text), with hybrid search and cited answers.
Voice: dictation with Whisper on the device or a provider's API, and reading answers aloud.
Sync: a GitHub repository, an S3-compatible bucket (AWS, R2, MinIO, B2, Garage), a folder, or an .aparte.zip archive. Only ciphertext and the keyslots leave the device.
Also: characters, side-by-side comparison of models, imports, exports, usage statistics, image generation, the share target, notification