Free & Open Source MIT License

SELF-HOSTED AI CHAT

A powerful AI chat interface that runs entirely on your machine. Connect any model, search the web, automate your browser, and never lose a conversation.

Ollama
Claude
OpenAI
OpenRouter
SEARCH the web for free
UNDERSTAND any document
REASON with any model
FAST local inference
SEARCH the web for free
UNDERSTAND any document
REASON with any model
FAST local inference

Built for humans. Powered by AI.

SURF — Chat Interface
Chat
Recent
Research task
Code review
Browser job
Tools
Analytics
Memory
SURF
Hey! What can I help you with today? I can search the web, read files, browse for you, or just chat.
Can you research the best local AI models right now?
SURF
Searching the web

Sure! Based on current benchmarks, here are the top picks for local deployment:

1. Llama 3.2 — Best reasoning & instruction following
2. Mistral 7B — Fast, efficient, great for coding
3. Qwen2.5 — Excellent multilingual support
SURF
Want me to open a browser and pull the latest leaderboard scores for each?
Ask anything, or type / for commands
100%
Local — nothing leaves your machine
$0
No subscriptions, no quotas
5+
AI providers supported
3
Platforms — Win, Mac, Linux

Everything you need.

01

Any AI Provider

Ollama (local), Claude, GPT, OpenRouter, or any OpenAI-compatible API. Switch provider mid-conversation with /provider.

02

Free Web Search

DuckDuckGo search with zero API key and no quota. Multi-source synthesis powered by Playwright. Works in both web UI and CLI.

03

Browser Agent

Vision-powered autonomous browsing. Give a task in plain English — SURF opens a real browser, takes screenshots, clicks, types, and navigates.

04

Persistent Memory

Global facts survive across all conversations. Session facts auto-extracted and deduplicated. Fully editable via sidebar or Memory modal.

05

Analytics Dashboard

Track messages, token counts, tokens/sec, response times, and model usage breakdowns across every conversation — rendered locally.

06

Vision & Images

Upload images in the web UI or attach via /image in the CLI. A dedicated vision model slot routes image messages automatically.

07

Conversation Summary

One-click AI summary of any conversation — structured into Topic, Key Points, and Outcome. Optionally save to memory with a toggle.

08

Skills System

Drop a Markdown file into skills/ to add new capabilities. Built-in: Web Researcher and File Reader. Enable/disable per-session.

09

Rich Terminal CLI

Full-featured CLI with slash commands, syntax-highlighted code, streaming output, multi-turn memory, and full parity with the web UI.

Power at your fingertips.

powershell — SURF CLI
python chat.py ███████╗██╗ ██╗██████╗ ███████╗ ██╔════╝██║ ██║██╔══██╗██╔════╝ ███████╗██║ ██║██████╔╝█████╗ ╚════██║██║ ██║██╔══██╗██╔══╝ ███████║╚██████╔╝██║ ██║██║ Model: llama3.2 Provider: ollama Memory: on   SURF❯ /search best Python frameworks 2025 [SEARCH] Querying DuckDuckGo... [DONE] 5 sources found · synthesising...   Based on current trends, FastAPI leads for APIs, Django for full-stack, and Flask for lightweight.   SURF❯ /web [LAUNCH] Web UI started → http://localhost:7777 Opening browser...

Running in 2 minutes.

1

Clone & Install

Run setup.bat on Windows or ./setup.sh on Mac/Linux. Creates virtualenv, installs deps, sets up Playwright.

2

Add a Model

Use Ollama locally for free, or add an API key for Claude, GPT, or OpenRouter. Works out of the box with zero key.

3

Start Chatting

Double-click start.bat, or run python chat.py. Web UI opens at localhost:7777.

One command away.

# Run setup (creates venv, installs deps, sets up Playwright)
.\setup.bat

# Launch (web UI at http://localhost:7777)
.\start.bat

# Or launch the terminal CLI
.\start.bat cli
# Make executable and run setup
chmod +x setup.sh && ./setup.sh

# Launch web UI
source venv/bin/activate
python chat.py
python -m venv venv

# Windows:
venv\Scripts\activate
# Mac/Linux:
source venv/bin/activate

pip install -r requirements.txt
playwright install chromium
python chat.py

Your AI.
Your machine.
Your rules.

No subscription. No data harvesting. No limits. Just a fast, private AI you actually own.