ZeroChat Docs
Español | English
Version 7.x

ZeroChat Help Center

Reference guide and technical documentation to get the most out of your portable, private AI client.

What is ZeroChat?

ZeroChat is a standalone, secure, and portable web client designed to interact with artificial intelligence models, orchestrate autonomous multi-turn agents, and search local document knowledge bases.

Unlike conventional web apps, ZeroChat is served as a static web application (zerochat.html) that runs directly on GitHub Pages or any static web server with no heavy dependencies, database backends, or proxies. All your personal data, chat histories, profile credentials, and indexed documents remain strictly on your machine within your browser.

Open ZeroChat (zerochat.html) Ready to use directly in your browser or connect with your local backend in zerochat.py.

Two Ways to Launch: Standalone Web vs. Local Backend

  • zerochat.html (Direct Web / PWA): Opens directly in any desktop or mobile browser (specifically optimized for Android as an installable PWA). By default it uses dark mode and, on mobile, keeps the conversation panel closed to prioritize the chat; both can be changed from the interface controls. Delivers commercial or local inference (WebLLM/WebGPU), in-browser RAG knowledge bases (Orama), Agent mode with a secure JS sandbox, and private IndexedDB storage without requiring Python, a terminal, or installation.
  • zerochat.py (Local with Python backend): Background local server (port 6388 with a secure ephemeral token). Extends the web app with host operating system control: reading and editing local project files, executing terminal commands, and connecting external MCP servers (Playwright, LSP, memory).
View full technical comparison & capabilities in Architecture →

Local server: recommended installation

Install the official executable from PyPI; it opens the GitHub Pages interface with a secure local session. PyPI and the direct download use the same ~/zerochat/ directory for state and MCP services.

pip install zerochat && zerochat

Direct alternative: download zerochat.py and run it with Python.

Download zerochat.py: zerochat.py

Run it directly (curl):

curl -sL https://albalday.github.io/zerochat/zerochat.py -o zerochat.py && python3 zerochat.py

Delete ~/zerochat/ to remove configuration, token, MCP services, and their dependencies without removing the installed or downloaded executable.

Self-learning project: ZeroChat was created to learn and experiment with AI and coding agents. View the details →

System Features & Modules

Explore each ZeroChat component below. Click any card to read its detailed technical guide:

Architecture & Portability
Static web application served over HTTPS, local persistence in IndexedDB, and total data sovereignty without server middleware.
Profiles & Providers
Connect with OpenAI, Ollama, OpenRouter, Claude, Gemini, and local Mirror with automated model and capability discovery.
Local WebLLM Inference
Execute models 100% inside the browser on your GPU via WebGPU, with catalog browsing, progressive downloading, and cache persistence.
Agent Mode & Tools
Autonomous multi-turn loop with sandboxed JavaScript execution in Web Worker, web search, page fetching, and interactive SVG charts.
Local Knowledge (RAG)
Ingest PDFs and documents into independent branches with vector and lexical indexing powered by Orama in the browser.
MCP Protocol
Local server with file and terminal tools, external MCP services, and permission controls.
Debugging & Interceptor
Debug console with filters, a Raw view, and an interceptor for outgoing messages.
Conversations & Branching
History, branches, search, and JSON, Markdown, or PDF export; JSON backup import.
Composer & Multimedia
Auto-expanding input textarea, keyboard shortcuts, and multi-format attachments (vision images, PDFs, source code, CSV, text).
Reasoning & Telemetry
Thinking effort control, agent checkpointing, and real-time monitoring of tokens, cache hit savings, and context window health.