dicedatalore/oolong

Oolong is a terminal-based chat client that lets you talk to several AI providers, including OpenAI, Anthropic, Google, and Ollama, from a single interface. You can switch between models mid-conversation without losing context, attach files and images, or pipe text in from other command-line tools. Chats live in memory only and are saved to disk only when you explicitly ask, which keeps things private by default. The project offers two main ways to work. Running oolong with no arguments opens a full interactive interface where you type, scroll, and manage conversations using keyboard shortcuts. Passing a prompt as an argument or piping input from another command produces a one-shot answer printed to stdout, useful for quick tasks like generating a commit message from a git diff. Both modes share the same model catalog, credentials, and provider settings. Configuration is optional but flexible. A config init command generates a starter TOML file you can edit to set a default model, accent colors, a simplified picker, reduced motion, and a custom system prompt. You can also define your own model entries, specifying provider, pricing rates, reasoning effort, and context window. If the config file is malformed or missing, the app starts with safe defaults and reports the issue in the interface rather than refusing to launch. API keys are stored in the operating system keychain rather than in a plain config file, and provider environment variables are also recognized. Local endpoints like Ollama or LM Studio require no key at all. The doctor command checks your local setup without contacting any provider, and OpenAI requests are sent with storage disabled. Saved transcripts are readable Markdown files that you can resume later with a --resume flag, though attachment contents are not embedded in the transcript itself. The README is thorough, covering installation for macOS, Windows, and Go, along with a full keybinding reference, CLI options, custom endpoint setup, and build instructions. It is written in Go and licensed under MIT.