Fox in the Box packages a Hermes Agent, the Hermes WebUI chat interface, persistent memory, and optional remote access into one desktop application. It runs a Docker container on your computer; the desktop app manages the container, and a browser-based wizard handles setup. You bring an API key for a provider such as OpenRouter or Anthropic, or point it at a local Ollama instance.
Messages go to the chosen provider; conversation history, files, and memory stay on the local machine. Memory uses mem0 with a Qdrant vector store, with embeddings computed on-device by a bundled nomic-embed-text model, so it works with any provider without an extra key. Optional Tailscale integration publishes the assistant over HTTPS to your other devices without port forwarding.
Most agent functionality comes from the bundled Hermes Agent and WebUI; Fox in the Box adds the desktop wrapper, Docker packaging, onboarding wizard, Tailscale setup, and an auto-update channel. Installers exist for Windows and macOS, with a script for Linux and macOS and a plain Docker command for advanced users.
Features
- Chat interface: streaming responses, tool-call progress, markdown, 26 slash commands, live steering, branching, and background tasks
- Persistent memory: mem0 plus Qdrant, on by default, with on-device embeddings and a panel to read, edit, and prune entries
- Multi-profile: separate Hermes environments with isolated config, sessions, memory, and a per-profile SOUL.md personality
- Skills and tools: bundled and user-authored skills, Brave Search via MCP, file I/O, terminal, web browsing, vision, TTS, transcription, and image generation
- Scheduled tasks: cron-style job builder that triggers a skill on a schedule
- Provider switching: add or change API keys from Settings without a restart
- Local models: auto-detects a running Ollama daemon and lists installed models for one-click switching
- Remote access: Tailscale integration with automatic HTTPS through Tailscale Serve
- Diagnostics: built-in log viewer and system diagnostics; all data under ~/.foxinthebox
Integrates with
Hermes WebUI
Browser interface for the Hermes Agent with chat, sessions, a workspace file browser, and profile and task controls
Mem0
Memory layer that extracts facts from conversations and retrieves them at inference so agents remember users across sessions
-
Ollama
Runs open large language models locally without an API key
Alternatives
Jan
Desktop app for running open-weight LLMs locally or connecting to cloud providers, exposing an OpenAI-compatible local API
LibreChat
Self-hosted ChatGPT-style chat platform unifying many AI providers with agents, MCP, and a code interpreter
Open WebUI
Self-hosted, offline-capable AI platform for Ollama and OpenAI-compatible models with RAG, tools, and multi-user access control
