Atomic Agent

Atomic Agent is a local AI agent for people who want an assistant to work with their browser, files and code on their own hardware. Its model and task loop can run offline, while sessions and memory stay on disk. It's free and open source under the MIT license.

It runs on Apple Silicon Macs, Linux and Windows, with local GGUF models served through llama.cpp. Small quantized models can use consumer GPUs or CPUs; an NVIDIA GPU isn't required. Local inference needs no account or API key. Cloud models through OpenRouter are optional, and choosing them sends model requests to that provider while the agent's state stays local. Web tasks and external integrations also need network access.

The agent can organize files after reading their contents, extract text locally from PDFs and Office documents, and operate a real browser for research. It saves cited notes in searchable memory and remembers context across sessions. For coding work, it can investigate repository issues, edit files, run shell commands and prepare fixes, with approvals for risky actions and before pushing changes.

Its llama.cpp backend constrains tool-call structure so small models produce valid actions. Cached context and compressed model memory reduce repeated computation and memory use during longer tasks. MCP connects additional tools, skills define repeatable workflows, and Telegram provides remote access and notifications for scheduled checks.