AgentCake
Your analytics cannot tell a person from an AI agent browsing on their behalf, so AgentCake records every visit as a session replay and classifies who was in control. Each session is labeled Human, Agent only, Human + agent, or Script bot, and agents are named, such as Claude in Chrome, ChatGPT, Grok Bot, or Muse, with a Verified or Likely label. Next to each replay, an activity timeline marks every hand-off between agent and human, and a Why this classification panel lists the reasons from four layers of evidence: name, verification, environment, and behavior. The dashboard counts sessions, visitors, page views, and active time for humans and agents, with breakdowns by path, source channel, device, and country, so you can see which pages agents read, where they stopped, and which agents visit most. A coding agent such as Claude Code or Codex can install it after you approve a plan with every monthly cost listed, connect sites on WordPress, Shopify, Wix, or Vercel, and verify the setup with a doctor command. Visits are sent only to your own server, no cookies, typed text, key presses, or visitor IP addresses are stored, and the dashboard opens only through an SSH tunnel. Pick the playful Cake style or the plain Minimal one. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.
Faster Chat
With Faster Chat, your team chats with OpenAI, Anthropic, Google, Groq, DeepSeek, or local Ollama models from one self-hosted window, and every conversation stays on your own server. The first account to register becomes admin, adds providers from a searchable hub that auto-discovers more than fifty services, picks which models people may use, and adds members or read-only users directly or through single-use invite links that expire after seven days. Members chat with streaming replies, Markdown, syntax-highlighted code, and LaTeX, attach files, send images to vision models, generate pictures through OpenAI, Replicate, or OpenRouter, and dictate by voice or have answers read aloud. Users can review or delete everything the memory feature has stored, opt a single chat out of it, and let web search cite its sources as clickable pills. Admins pull Ollama models from the admin panel with live download progress instead of a command line, and anyone can import a ChatGPT export or switch among fifteen-plus color themes, custom fonts, light and dark modes, or a focus mode. White labeling swaps in your organization's app name and sidebar icon, API keys stay encrypted, and one Docker Compose command starts the stack, with optional HTTPS through Caddy. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.