开源AI助手平台

Wordmark

An open source AI assistant platform

Overview

Wordmark is a client-side AI chat for OpenAI/xAI Responses APIs and local LM Studio or Ollama servers. It supports tool/function calling, TTS, themes, and fully local storage — no backend required.

Screenshots

Desktop chat with streaming replies, a collapsible reasoning panel, and inline code — Aurora theme, running locally on google/gemma-4-12b-qat via LM Studio.

Configure everything from the settings panel — provider and model, tool calling, personality, and themes:

Fully responsive on mobile, with a range of built-in color schemes:

Docs:

• Getting Started
• Overview
• Services & Models
• Tool Calling
• Skills
• Party Mode
• Autonomous Work
• Streaming
• Memory
• Security
• Storage
• UI & UX
• Docker
• Electron App
• Troubleshooting

Features

• Providers — OpenAI Responses (hosted), xAI Grok (Responses-compatible), and local LM Studio or Ollama servers (Services & Models)
• Tool calling — built-in weather, provider web + X search, Code Interpreter, image generation, optional file search over vector stores (OpenAI), and custom MCP servers (Tool Calling)
• Documents — attach files or folders in any text/code format plus PDF, Office, OpenDocument, and ebooks; hosted providers take the file itself, local providers extract and search it in-browser with embeddings so nothing leaves your machine (Documents & Attachments)
• Skills — uploadable SKILL.md instruction packages the assistant loads on demand to specialize for a task; ships with example skills (Skills)
• Streaming & reasoning — dedicated reasoning panel, rich tool timelines, inline code previews, automatic image capture (Streaming)
• Party mode — autonomous multi-character group chat over your chosen provider; build a cast and scenario, then interject any time without pausing (Party Mode)
• Prompt queuing — type while a response is still streaming and the message reaches the assistant mid-answer: the reply in flight is interrupted and resumed with your message added, keeping every word already written (UI and UX)
• Autonomous work — opt in and a message becomes a goal: the assistant queues its own next steps and keeps taking turns until the work is done, inside a turn budget you set, with pause/resume/stop controls (Autonomous Work)
• TTS — OpenAI (13 voices) and xAI (5 voices) providers, optional autoplay, per-message controls, audio cached locally
• UX — themes, responsive layout, syntax highlighting, markdown, image gallery (UI & UX)
• Local-only storage — conversations, images, and audio via IndexedDB; keys stay in the browser (Storage)
• Memory — local, FIFO-limited memories appended to the system prompt (Memory)

Quick Start

git clone https://github.com/h1ddenpr0cess20/Wordmark.git
cd Wordmark
npm install
npm run dev          # dev server on http://localhost:3000
# npm run dev:https  # HTTPS — needed for some APIs, TTS, and geolocation

Wordmark builds with Vite. Use npm run dev for development or npm run build to produce a static bundle in dist/ (serve it with npm run preview). Opening index.html straight from the filesystem no longer works — the app must be served by the dev server or built first. See Getting Started.

1. In Settings → API Keys, add your OpenAI/xAI keys. Keys and URLs are stored locally.
2. Choose a provider and model in Settings → Model.
3. Type a message and send.

Local Models

• LM Studio — run the server (default http://localhost:1234), set the base URL in Settings → API Keys, then select LM Studio in Settings → Model (LM Studio guide)
• Ollama — run the server (default http://localhost:11434), set the base URL in Settings → API Keys, then select Ollama in Settings → Model
• Both take an optional API key in the same settings block, for a server started behind one; leave it blank and requests stay unauthenticated

Note: Chrome may prompt you to allow local network access. This is only used to connect to local LM Studio/Ollama servers.

HTTPS & Docker

HTTPS is recommended for full functionality — quick steps in Getting Started. Full Docker/Compose instructions and SSL options in the Docker guide.

# Pull from Docker Hub and run
docker run --rm -p 8080:80 h1ddenpr0cess20/wordmark:latest

Or build from source:

docker build -t wordmark:latest .
docker run --rm -p 8080:80 wordmark:latest

Architecture & Development

• Architecture — high-level structure
• Security & Storage — data handling
• UI & UX — layout and design
• Development & CONTRIBUTING — developer guide

Common tasks:

• Add tools — register tool metadata in src/ts/services/api/staticTools.ts, implement handlers (see src/ts/services/weather.ts), and tune provider quirks in src/ts/services/providers.ts — Tool Calling
• Adjust models/providers — edit src/config/config.ts — Services & Models
• Themes and styling — src/css/themes/**, src/css/components/**

Usage

• Enable Tools in Settings to allow function calls for weather, web search, file attachments, and any MCP servers you connect (Tool Calling)
• Manage conversations, images, and audio locally via History and Gallery
• Use TTS for spoken responses — configure provider and voice in Settings → TTS

Policies & Notes

• Privacy/Security — client-side only; no tracking (Security)
• Troubleshooting — common issues and tips (Troubleshooting)
• Not a Companion — philosophy and boundaries (Not a Companion)

License

MIT — see LICENSE