#6565th of 107 in Coding agents and AI editors

atomic-agent

Local-first terminal agent that browses, edits files and runs approved commands

Stars
3.4k
License
MIT
Last commit
Oct 2026
Last release
Oct 2026
Language
TypeScript
Stars, last 26 weeks+3,423
21 Apr 2026: 3 stars28 Apr 2026: 3 stars5 May 2026: 5 stars12 May 2026: 9 stars19 May 2026: 14 stars26 May 2026: 62 stars2 Jun 2026: 68 stars9 Jun 2026: 72 stars16 Jun 2026: 77 stars23 Jun 2026: 81 stars30 Jun 2026: 85 stars7 Jul 2026: 432 stars14 Jul 2026: 452 stars21 Jul 2026: 489 stars28 Jul 2026: 865 stars4 Aug 2026: 985 stars11 Aug 2026: 1,377 stars18 Aug 2026: 1,922 stars25 Aug 2026: 2,132 stars1 Sept 2026: 2,217 stars8 Sept 2026: 2,282 stars15 Sept 2026: 2,314 stars22 Sept 2026: 2,368 stars29 Sept 2026: 2,432 stars6 Oct 2026: 2,919 stars13 Oct 2026: 3,296 stars20 Oct 2026: 3,426 stars

3 on 21 Apr 20263,426 on 20 Oct 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

Atomic Agent is a TypeScript CLI/TUI agent that keeps its control loop, memory and sessions on your disk. It drives a browser via playwright-core, edits files, runs approval-gated shell commands, and connects to MCP servers. It targets a local llama.cpp server (managed or your own) with grammar-constrained tool calls, and also supports cloud providers and a Fusion mode that splits planning and work across models.

Who it is for: Engineers running small local models for long, tool-heavy agent tasks

Strengths

  • Runs on local llama.cpp models with GBNF-constrained tool calls; no API cost locally
  • Reports GAIA L1 69.8% vs Hermes 58.5% on the same local model
  • Memory in SQLite/FTS5 and plain-file state, with traces you can inspect
  • Imports sessions and skills from Claude Code, Codex, Pi, Hermes and OpenClaw

Weaknesses

  • Developer preview: APIs, commands and config are still changing
  • No Intel Mac build; Windows on ARM runs x64 under emulation
  • Install is a curl-to-shell script; no Docker or compose files
  • Benchmark covers 53 tasks, with different agent versions per model row

What it needs

  • GPU optional
  • Needs llama.cpp (llama-server), Chrome/Edge/Brave/Chromium (browser tools)
  • Models: GGUF via llama-server, OpenAI-compatible, OpenRouter, Gemini, Anthropic

Also in Coding agents and AI editors

See all 107
Also in Coding agents and AI editors
RankProjectScore
1claw-codeRust CLI coding agent harness using Anthropic and OpenAI-compatible APIs195.2k starsMIT83 out of 100
2ArchonYAML workflow engine that runs AI coding agents through fixed dev steps23.7k starsMIT83 out of 100
3Open SWEAgent that turns issues into pull requests, reviews PRs and monitors CI11.1k starsMIT81 out of 100
#4oh-my-openagentTerminal coding agent with multi-model orchestration and git-backed memory70.2k starscustom license80 out of 100
#5gooseRust AI agent with desktop app, CLI and API, extended through MCP55.5k starsApache-2.080 out of 100