🦒 Giraffe AI

The lightest, most powerful AI chat client. Zero installation, zero dependencies — start and chat.

Giraffe AI is a featherweight AI chat client that runs directly in your browser. Nothing to install, nothing to configure, no runtime to download: double-click the launcher and you are chatting with your AI models in seconds. It is the fastest, simplest and most resource-efficient way to talk to local or cloud AI models — an immediate, professional chat experience that barely touches your system.

No installation · No dependencies · No setup · Start and go.

---

Why Giraffe AI?

⚡ Instant start	No installer, no package manager, no build step. Launch and chat immediately.
🪶 Ultra-lightweight	A single HTML file with near-zero RAM and CPU usage — it does not occupy resources.
🔌 Any AI model	OpenAI, DeepSeek, Groq, OpenRouter, Anthropic (via OpenRouter), local models via Ollama, or any OpenAI-compatible endpoint.
🚀 Zero install	Ships with ready-made launchers for Windows, Linux and macOS.
🔒 Private by design	Conversations and settings live in your browser's local storage; your API keys never leave your machine.
💬 Professional chat	Chat history, Markdown rendering, code blocks with one-click copy, reasoning/thinking display, regenerate answers, fine-grained settings (temperature, max tokens, top-p).
🌍 Multi-language UI	English, Italian, French, German, Spanish, Russian.

Quick Start

Windows — double-click start.bat

Linux / macOS — run ./start.sh

The launcher starts a tiny local server, opens the app in app-mode, and proxies /v1/* requests to a local OpenAI-compatible model server (e.g. llama.cpp, Ollama, LM Studio) on 127.0.0.1:8787 — so local models work with no CORS workarounds and no API key.

Cloud providers (OpenAI, DeepSeek, Groq, OpenRouter, …) are configured in seconds from the built-in presets: pick a preset, paste your API key, done.

Alternatively, open index.html directly in any modern browser. Cloud APIs can be used immediately; the launcher is only needed for local model servers.

Supported Providers

Provider	Type	Preset	Native format
OpenAI (GPT-4o, o3-mini)	Cloud API	✅	OpenAI-compatible
DeepSeek (chat, R1 reasoner)	Cloud API	✅	OpenAI-compatible
Ollama (local)	Local model	✅	Ollama /api/chat
Groq (Llama 3.3)	Cloud API	✅	OpenAI-compatible
OpenRouter	Cloud API	✅	OpenAI-compatible
Anthropic Claude (via OpenRouter)	Cloud API	✅	OpenAI-compatible
Anthropic Claude (native)	Cloud API	✅	Anthropic Messages API
Gemini 2.0 Flash (native)	Cloud API	✅	Gemini API (inlineData)
Any OpenAI-compatible endpoint	Cloud / Local	Custom	OpenAI-compatible

Features

• Multiple providers & models per workspace, switchable from the sidebar
• Chat history per provider, persisted locally
• Markdown rendering with syntax-highlighted code blocks and copy button
• Reasoning / thinking display for reasoning models (e.g. DeepSeek R1, o3-mini)
• Regenerate & copy for any assistant message
• Chat settings — system prompt, temperature, max tokens, top-p (global or per chat)
• Light / dark theme and 6 UI languages
• Installable as a PWA (app-mode / standalone window)
• Streaming responses with stop-generation control
• Multi-file attachments — attach several files at once (or one at a time) to any message; the original bytes are uploaded and converted to Markdown server-side by the connected backend (AgentBridge /v1/files, or any OpenAI-compatible file endpoint)

Giraffe AI vs. the Big Five

Compared with the five most popular AI chat platforms — ChatGPT, Claude.ai, Gemini, Open WebUI and LM Studio — Giraffe AI stands out as the simplest and lightest way to use LLMs:

	🦒 Giraffe AI	ChatGPT	Claude.ai	Gemini	Open WebUI	LM Studio
Installation	None — a single HTML file	App + account	App + account	App + account	Docker / Python environment	Desktop installer
Dependencies	Zero	Vendor app	Vendor app	Vendor app	Python + many packages	Bundled runtime
Time to first chat	Seconds — start and go	Sign-up + sign-in	Sign-up + sign-in	Sign-up + sign-in	Long setup	Download + install
Resource usage	Minimal (one browser tab)	Heavy browser app	Heavy browser app	Heavy browser app	Server + database always running	Multi-GB install
Models supported	Any — OpenAI, DeepSeek, Groq, OpenRouter, Anthropic, Ollama, custom	OpenAI only	Anthropic only	Google only	Any	Local only
Local models	✅ built-in proxy, no CORS hacks	❌	❌	❌	✅ (adds complexity)	✅
Privacy	Data stays in your browser (localStorage)	Stored on servers	Stored on servers	Stored on servers	Self-hosted	Local
Offline	✅ (with local models)	❌	❌	❌	✅	✅
Cost	Free	Subscription	Subscription	Subscription	Free (self-hosted)	Free
App-like experience	✅ optional PWA install	✅	✅	✅	❌	✅

What makes Giraffe AI different

• Zero-install innovation — the entire client is one self-contained HTML file: no framework, no Node.js, no Python required to use it. Every other client above needs an installer, a package manager, or a full runtime environment.
• Instant by design — no account, no onboarding, no build step. Double-click and you are talking to an LLM.
• Provider-agnostic — one lightweight UI for cloud APIs and local models. No vendor lock-in, no model ceiling.
• Near-zero footprint — nothing runs in the background, no server daemon, no database. It occupies essentially no resources.
• Private by design — API keys and chat history never leave your machine; the only network calls go to the model providers you explicitly configure.
• Smart same-origin proxy — the launcher embeds a tiny reverse proxy (/v1/* → 127.0.0.1:8787) so local models work without CORS workarounds, browser flags, or extensions.

Giraffe AI vs. other API clients

There is a whole family of apps that, like Giraffe AI, are pure clients: you bring your own API key and connect to whichever provider you want. Compared with the most popular of them — Chatbox, Jan, AnythingLLM, NextChat and LibreChat — Giraffe AI is the only one that delivers the same power with no application chassis at all:

	🦒 Giraffe AI	Chatbox	Jan	AnythingLLM	NextChat	LibreChat
Type	Web client — single HTML file	Desktop app (Electron)	Desktop app (Electron)	Desktop app / Docker	Self-hosted web app (Next.js)	Self-hosted web app (Next.js)
Installation	None — open the file, start and go	Installer	Installer	Installer or Docker	npm + Node.js	Docker or npm + MongoDB
Dependencies	Zero	Bundled Electron runtime	Bundled Electron runtime	Node.js + vector DB	Node.js + npm packages	Node.js + MongoDB
Resource usage	Minimal — one browser tab	Hundreds of MB (Electron)	Hundreds of MB (Electron)	Server + database	Node.js server	Node.js + database servers
Time to first chat	Seconds	Install + setup	Install + setup	Setup + configuration	Build + deploy	Full-stack setup
Portable single file	✅ — copy it anywhere, it runs	❌	❌	❌	❌	❌
Models	Any OpenAI-compatible + local	Many (by key)	Local + cloud (by key)	Many + RAG	Many (by key)	Many (by key)
Offline	✅ (with local models)	✅	✅	✅	✅	✅
Cost	Free	Freemium	Free, open source	Free, open source	Free, open source	Free, open source

These apps share Giraffe AI's philosophy — one UI, any provider — but they carry a heavy chassis: an Electron runtime, a Node.js stack, or a Docker + database setup that must be installed, updated and maintained before you can chat. Giraffe AI delivers the same capability as a single portable HTML file that runs anywhere a browser runs: nothing to install, nothing to update, nothing running in the background.

How It Works

Giraffe AI is intentionally simple:

• index.html — the entire chat client, self-contained (HTML + CSS + JS), no external libraries.
• start.bat — Windows launcher: a built-in PowerShell HttpListener on port 8000 that serves the app and proxies /v1/* to 127.0.0.1:8787 (same-origin, so no CORS).
• start.sh — Linux/macOS launcher: uses python3 (with the macOS built-in Ruby as fallback) for the same role.
• All data is stored in the browser's localStorage — nothing is sent anywhere except the model APIs you configure.

Auto-configuration (--provider)

Start the launcher with a provider configuration and the client registers it (if not already present, same name + endpoint + format) and selects it as active — so AgentBridge and similar hosts can open a ready-to-chat window with no manual setup:

# Linux / macOS — raw JSON (URL-encoded automatically by the launcher):
./start.sh --provider '{"name":"AgentBridge","format":"openai","model":"default-agent","endpoint":"http://localhost:5290/v1/chat/completions"}'

# Windows — the JSON is passed base64url-encoded (no padding; safe on the cmd.exe command line):
start.bat --provider eyJuYW1lIjoiQWdlbnRCcmlkZ2UiLCJmb3JtYXQiOiJvcGVuYWkiLCJtb2RlbCI6ImRlZmF1bHQtYWdlbnQiLCJlbmRwb2ludCI6Imh0dHA6Ly9sb2NhbGhvc3Q6NTI5MC92MS9jaGF0L2NvbXBsZXRpb25zIn0

The launcher appends ?provider=<url-encoded JSON> to the opened URL; on load the client decodes it, registers/selects the provider and clears the query string (a later refresh does not re-apply it). Supported fields: name, format (openai / ollama / anthropic / gemini), model, endpoint, apiKey, temperature, maxTokens, topP, reasoningEnabled, systemPrompt, attachments, textLimit, id.

⚠️ Security note: when an apiKey is included it is visible in the URL and stored in the browser's localStorage. Keep --provider auto-config for local/trusted hosts and avoid passing secrets on shared machines.

File Attachments

Attach files to any message via the paperclip button: several files can be selected at once (Ctrl/Shift-click) or added one at a time; the selected files are listed under the input and can be removed individually. Attachment support is per-provider configurable:

• Text files are always available. A large list of textual extensions (.txt, .csv, .md, .yaml, .py, .js, .ts, .json, .xml, config/code files, Dockerfile, Makefile, README, …) is always attachable: the client reads them in the browser (FileReader) and embeds their content in the message (truncated to the provider's Text Char Limit / File, default 30 000 chars). No backend needed.
• Images & documents need the provider flag. The provider form has a toggle "Allow image/document attachments" (on by default). When off, only text files can be attached. When on:
  • Images are sent inline (base64) using the provider's native format — image_url for OpenAI-compatible, source.base64 for Anthropic, inlineData for Gemini, images[] for Ollama.
  • Documents (PDF/DOCX/XLSX/…) are uploaded to the backend POST /v1/files (e.g. AgentBridge → AllToMarkdown → file_id); the chat request includes the file_ids in the request body (file_ids) as context. If the upload fails, the send is blocked with a clear error instead of silently losing the content. Anthropic native also supports documents inline (base64) and falls back to it when the provider has no /files endpoint.
• The text content of each textual file is embedded once: it is shown in the message bubble and sent in the API payload a single time (no duplication).

Provider presets set sensible defaults; the toggle and the text limit are editable per provider in the provider modal. All of this is implemented inside index.html — consistent with the single-file philosophy.

Tests

tests/test_attachments.js runs the attachment logic headlessly (no browser needed): requires Node.js and executes the pure functions extracted from index.html (isTextualFile, uploadFiles, buildMessageWithAttachments, buildAPIMessagesWithFiles — including the per-format payloads for OpenAI, Ollama, Anthropic and Gemini, and the clear-error path when a binary upload is not supported):

node tests/test_attachments.js

Contributing

The goal of Giraffe AI is to be a complete, functional and powerful chat client built on a radically minimalist foundation. Every contribution must preserve these principles:

• No installation for the user — the app must keep running as a single, self-contained index.html opened directly in any modern browser. No build steps, package managers, or runtime requirements for end users.
• Zero dependencies — no external libraries, frameworks, or CDN imports. Everything stays hand-rolled vanilla HTML/CSS/JS, exactly as it is today.
• Minimalist under the hood — prefer simple, readable code over abstractions. Add a feature only if it does not require installing complex components or moving the project to a full toolchain.
• Powerful by value, not by weight — streaming, reasoning display, multi-provider support and chat history belong in the client; heavy infrastructure does not.
• The launcher stays optional — start.bat / start.sh only serve the page and proxy local models for convenience; they must never become a requirement to run the app.

If you want to contribute, open an issue first to discuss the change — especially anything that would introduce a dependency or an installation step. Pull requests that keep the project dependency-free and single-file are always welcome.

Search Keywords

If you are looking for any of the following, this project is for you: lightweight AI chat client, no-install AI chat, zero-dependency LLM client, portable AI chatbot, minimal resource usage chat, browser-based AI chat, local LLM web client, OpenAI-compatible chat UI, lightweight ChatGPT alternative, offline-friendly AI assistant.

---

Giraffe AI — simple, powerful, immediate. No installation, no dependencies. Start and go.