Dive

Dive is a desktop AI agent for people who want a language model to work with files, shell commands and connected tools. It supports models with function calling and acts as a host for Model Context Protocol (MCP) servers.

• Model choice: Use Ollama for a local LLM, connect to ChatGPT or Anthropic, or choose an OpenAI-compatible model, with support for model switching and multiple API keys.
• Built-in tools: The agent can make web requests through Fetch, read and write files through File Manager, and execute commands through Bash.
• MCP connections: Connect local or remote MCP servers and use an installer agent to help set up additional servers; optional OAPHub.ai integration provides access to managed servers in the cloud.
• Agent controls: Enable or disable individual MCP tools and set custom system prompts to adjust the agent's available actions and behavior.
• Chat input: Attach file references through path search, keep unfinished message drafts and use the interface in multiple languages.

Dive is open source under the MIT license and runs on Windows, macOS and Linux. Electron builds cover all three platforms; Tauri builds are available for Windows and Linux. The desktop app and built-in file and command tools run on your machine. Cloud models send requests to their providers, and OAPHub.ai's managed MCP servers run remotely and require an account. MCP server authentication is unstable and may need repeated authorization.