Otis AI Agent

Otis is an open-source personal AI agent with a native desktop app, an OpenTUI terminal interface, and a headless mode for scripts and CI. Use it for everyday work, from planning a trip and drafting an email to exploring a codebase and fixing a bug. Otis can inspect files, edit code, run commands, search the web, and delegate focused work to subagents.

Otis makes local models easy to run. During setup it recommends a model for your hardware, downloads it, and runs it through llama.cpp. With a local model, Otis works fully offline: there is no account, telemetry, hosted control plane, or cloud synchronization, and your configuration and history stay on disk.

If you have additional NVIDIA hardware on your network, Otis can connect to an NVIDIA PAIR cluster. When you want a larger open-weight model, it can also connect directly to Fireworks with your own API key; Fireworks documents Zero Data Retention for open-model inference by default.

Why Otis?

• Your machine, your state. Configuration, sessions, tool activity, diffs, and usage statistics stay local.
• Your choice of open model. Use an Otis-managed GGUF, let PAIR route across your computers, or use Fireworks serverless when you want hosted performance.
• Desktop, terminal, or automation. Use the native desktop app, the OpenTUI interface, or headless mode with the same agent behavior and local sessions.
• Focused delegation. Otis can hand off exploration and research to subagents whose work remains inspectable.
• Direct provider connections. Hosted inference goes directly to Fireworks with your API key; web access goes directly to Parallel's Search MCP.
• Inspectable history. Append-only JSONL sessions preserve messages, tool cards, diffs, and provider-reported usage.