Hope Agent
A desktop AI agent for macOS or self-hosted Docker servers, with persistent memory, background tasks, Ollama models and direct cloud provider connections.
Hope Agent is a local-first personal AI agent for people who want ongoing project work to continue across conversations and devices. It runs as a macOS desktop app or a self-hosted service on a NAS, home server or cloud VM, with browser access. It's open source under MIT. Linux and Windows desktop support is experimental.
The agent can pursue goals in the background, split work between agents and check results against completion criteria. You can pause and resume tasks, review progress or schedule recurring work. With approval, it can operate the macOS desktop and browser, edit files, run commands and use MCP tools. Docker sandboxing limits where risky commands and file writes can act.
Memory carries project context and preferences across sessions, and the agent can turn completed work into reusable skills for review. Knowledge Space lets you and the AI work on Markdown notes together, including existing Obsidian vaults, with search, backlinks and traceable sources. Incognito conversations disable persistent memory and remove session data when they end.
Design Space creates websites, mobile prototypes and presentations from prompts or reference images. It includes visual editing, version history and exports such as HTML, PDF and PPTX. Designs can connect to a real repository for implementation, while sessions and task state can continue through Telegram, Discord, Slack or Feishu.
Ollama supplies local LLMs and embedding models without an account or API key. Cloud model requests go directly to the selected provider; configuration, conversations, memory and attachments stay on the machine hosting the agent by default.
