kevrojo/dulus

Dulus is a command line AI agent that lets you use AI models without needing an API key, credit card, or subscription. The core idea: instead of paying for API access, Dulus opens your web browser, captures your active session on services like Gemini, Claude.ai, Claude Code, Kimi, Qwen, or DeepSeek, and then treats that browser session as if it were a proper API. You can then issue commands from your terminal and have the AI read files, edit code, run shell commands, search your codebase, browse the web, and create commits on your behalf. Beyond the browser session approach, Dulus also supports over 100 paid AI providers through a routing layer called LiteLLM, plus local models running on your own machine through Ollama. It includes features like multi model debate sessions where several AI models discuss a problem together, voice input and output, a semantic memory system called MemPalace, local text extraction from images, a plugin system for adding Python libraries on the fly, an embedded sandbox operating system, and support for 34 languages. The whole thing is written in roughly 12,000 lines of readable Python with no build step required, just install and run. The README notes it is maintained by a single contributor and mentions a community launched Solana token associated with the project, which the creator says was not created by them. Installation is done through pip or a one line shell script. You would reach for Dulus if you want an autonomous AI coding assistant but don't want to pay for API access, or if you want a single tool that can wire together multiple AI services from one terminal prompt. It is released under the GPLv3 license.