AIlice
A self-hosted AI agent for research, coding and system tasks. Works with local LLMs or cloud APIs on Ubuntu, macOS and Windows through Docker or WSL.
AIlice is an open source AI agent that runs on your own machine and handles tasks spanning research, software development and system administration. It's for people who want an assistant that can act on files, browse the web and execute code. Its Interactive Agents Call Tree architecture splits complex requests among agents created for the task, then combines their results.
The browser interface supports voice conversations, file uploads and downloads, and displays images, video, audio, LaTeX formulas and highlighted code. AIlice can read scholarly papers, produce literature reviews and write and run scripts. You can interrupt a task to guide the agent as it works. MCP support connects it to external tools, and it can write and load extension modules to add capabilities.
You can use local LLMs through LM Studio or Ollama with LiteLLM, or connect an OpenAI-compatible inference server. Cloud choices include Anthropic, OpenAI, Mistral and Groq. Different agents can use different models, so a task can combine local inference with commercial services. Local models run on your hardware; requests to cloud models go to their providers and require API credentials. kragent.ai is a separate hosted service built on AIlice.
AIlice uses the MIT license. It runs on Ubuntu and macOS, with Docker or WSL recommended for Windows. Native Windows command-execution tools are incomplete and Windows testing is limited. Chrome handles web browsing. Local inference can be demanding: the project's hardware guidance calls for large models and two RTX 4090 GPUs with 48 GB of combined VRAM for effective task execution. Using cloud models avoids that local GPU requirement.