AtomicBot-ai/atomic-agent

Atomic Agent is a local-first AI agent. Runs open-weight models on your own machine via llama.cpp.

What it solves

Atomic Agent is a local-first AI agent designed to run on a user's machine to perform complex, multi-step tasks without the high costs or privacy concerns of cloud-hosted services. It addresses the challenge of using small local models for long-term work by implementing a context-frugal loop that prevents the model from being overwhelmed by growing conversation history.

How it works

The agent operates on a control loop where the model decides on an action (via grammar-constrained JSON tool calls), executes the action, and then compresses the results rather than pasting them back into the prompt. It leverages a custom TurboQuant llama.cpp backend to increase throughput and reduce memory usage via KV-cache quantization and speculative decoding.

To maintain efficiency, it externalizes state—such as sessions, memory, and tasks—into local SQLite databases and files, meaning the prompt only contains compact pointers that are recalled only when necessary. It can run in "Fusion" mode, where a cloud model acts as an orchestrator while local models handle the bulk of the execution.

Who it’s for

Developers and power users who want a private, local-first AI agent capable of driving a browser, editing files, and running shell commands on their own hardware using quantized local models or a hybrid cloud-local setup.

Highlights

  • Local-First Architecture: All state, memory, and control loops live on the user's disk.
  • Broad Tool Surface: Capable of web browsing (Playwright), filesystem operations, shell command execution, Git/GitHub integration, and document extraction.
  • ** uma MCP Support:** Connects to external Model Context Protocol (MCP) servers to expand its tool registry.
  • Advanced Memory System: Uses a hybrid recall system with profile facts, episodic notes, and distilled lessons stored in SQLite.
  • High Performance: Optimized for small local models using TurboQuant for 30-50% higher throughput.
  • Fusion Mode: Allows a high-capability cloud model to plan and delegate tasks to a pool of local worker models.

Related

  • Project
    AtomicBot-ai/Atomic-Chat

    Atomic Chat is a desktop/mobile app that runs open‑weight LLMs locally and exposes them via an OpenAI‑compatible HTTP server (`http://localhost:1337/v1`). It supports multiple back‑ends (llama.cpp, a TurboQuant‑enhanced fork, and Apple‑Silicon MLX‑VLM), speculative decoding tricks for speed, cloud‑model fallbacks, and one‑click integration with many open‑source coding agents. The UI handles model download, engine selection, and API routing, making private on‑device inference easy for developers and researchers.

  • Project
    QwenAudio/qwen-audio-agent

    A realtime voice runtime that enables AI agents to maintain continuous, full-duplex conversations while simultaneously executing background tasks and tool calls.

  • Project
    zhayujie/CowAgent

    CowAgent is an open‑source, self‑hosted AI assistant that plans tasks, runs tools, keeps a personal knowledge base, and can operate as a team of cooperating agents. It supports many LLM providers, multimodal input, dozens of chat channels (Telegram, Slack, WeChat, etc.), a plug‑in skill system, and a three‑tier memory with nightly “Deep Dream” distillation. Install via a one‑line script or Docker, manage everything through a web console, and extend the system through the public Skill Hub or custom tool integrations.

  • Project
    AgentiLoop/Agent

    AgentiLoop Agent is a native macOS app that lets you converse with any major LLM and have the model autonomously perform real‑world tasks—coding, GUI automation, shell commands, web browsing, voice control, and custom Swift scripts—through a self‑verifying task loop and a rich toolbox of built‑in tools.

  • Project
    Fosowl/agenticSeek

    A private, local AI assistant that autonomously browses the web, writes code, and manages files while keeping all data on the user's device.