← Back to Discover

Atomic Agent

A local-first AI agent that runs small quantized models on your own machine through llama.cpp, with tools, memory and scheduling, driven from a terminal, CLI, HTTP API or chat apps.

Autonomous agentsLocal modelsCommand-line tools
Visit website ↗

What it does

Atomic Agent is an AI assistant you run yourself. It uses small quantized language models through llama.cpp on your own hardware, so files and conversations stay on your machine by default. No account is needed, and an NVIDIA GPU is not required. The agent can browse the web, read and edit files, run shell commands, extract text from documents, remember things across sessions and schedule background tasks. Shell commands, file writes and HTTP requests pause for approval. Skills and MCP servers can be added, and cloud model providers can be used as well. You drive it from a terminal UI, a CLI, an OpenAI-compatible HTTP API, or chat apps such as Telegram and Discord. The docs describe it as a Developer Preview, tracking release v0.6.5, with commands and configuration still changing. Prebuilt binaries cover macOS on Apple Silicon, Linux x64 and arm64, and Windows x64; Intel Macs are not currently built. Anonymous usage analytics and crash reports are on by default and can be switched off in config.json.

Capabilities

  • Runs local quantized models through llama.cpp with no per-token fees
  • Tools for the browser, files, shell, git, documents and vision
  • Memory kept in a local SQLite store across sessions
  • Skills and MCP server support
  • Scheduling through cron, intervals and webhooks
  • Terminal UI, CLI, OpenAI-compatible HTTP API, Telegram and Discord

When to use it

  • Running an AI agent privately on your own hardware
  • Avoiding per-token costs by using local models
  • Automating background tasks on a schedule
  • Experimenting with a hackable, MIT-licensed agent