Atombot – tiny personal assistant for local models and GPT‑5.4

Name: Atombot – tiny personal assistant for local models and GPT‑5.4
Availability: InStock
Author: gwangee

by gwangee·Mar 8, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidBig BrainCozy

Tiny footprint (~500 lines vs 400k OpenClaw), but local AI assistants are crowded.

Strengths

•99.9% smaller codebase than OpenClaw with feature parity (skills, memory, reminders)
•Provider-agnostic onboarding auto-detects Ollama, LM Studio, Codex, GPT cleanly
•Telegram gateway with allowlist access and searchable daily history logs

Weaknesses

•Local LLM assistant category saturated (OpenClaw, nanobot, Jan, Ollama WebUI exist)
•No clear differentiation beyond code simplicity—features are baseline now

Post Description

Atombot(repo: https://github.com/daegwang/atombot) is a tiny self-hosted personal AI assistant I built because I wanted something I could understand and extend end-to-end.

Core functionality is ~500 lines (vs ~400k in OpenClaw). It includes: - Persistent memory with searchable daily history logs - Telegram gateway (allowlist-based access - One-time + recurring reminders - Skills system (OpenClaw SKILL.md compatible) - Provider-first onboarding that auto-detects Ollama / LM Studio / Codex

Atombot also supports GPT‑5.4 via Codex CLI, so you can select easily during onboarding.

Feedback welcome. Stars appreciated!