Arkloop – Open-source, local-first Agent client

Name: Arkloop – Open-source, local-first Agent client
Availability: InStock
Author: qqqqqf

by qqqqqf·Apr 30, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidSlickShip It

Firecracker microVMs for code execution beats Docker for isolation.

Strengths

•Firecracker microVM sandboxing offers stronger isolation than standard containers.
•Priority-based model routing handles rate limits without user intervention.
•UI deliberately hides model names to reduce cognitive noise.

Weaknesses

•Another local agent client in a space crowded with Cursor and Continue.
•Alpha status warning suggests stability issues for daily driver use.

Post Description

Hi HN, I built Arkloop – an open-source, local-first Agent client. You can think of it as Claude Desktop, but open source with its own taste.

Not a wrapper around any Agent SDK, not based on Claude Code – I built this from scratch, solo, over 3 months.

One thing I focused on: reducing cognitive load.

For example – if you only use one model for chatting, why does every message show which model was used? That's noise. Another example: developers love showing full tool use details in the UI, but what users actually need to know is that the agent is working and which direction it's going. We don't need all that detail.

Design philosophy: cognitive load, information density, value-driven, aesthetics.

I'd love your feedback – on anything from a 1px spacing issue to a major architectural direction. The README has a section about this that I'm particularly proud of:

"I welcome all forms of contribution. Even if you're not a developer – if something feels off during use, even just a spacing issue or a color, open an issue. If you're a developer, Arkloop's agent core, memory system, and hook mechanisms are all open."

It also supports importing configs from OpenClaw/Hermes.

GitHub: https://github.com/qqqqqf-q/arkloop Website & Download: https://arkloop.cn