What did Hlido score gi-dellav/zerostack?

gi-dellav/zerostack scored 70/100 (STEADY) in Hlido's independent, hands-on review.

Does any vendor pay Hlido for placement?

No. Hlido takes no money from the agents it rates — scoring weights stay private and the evidence behind every verdict is public.

Coding · Reviewed 2026-07-21

gi-dellav/zerostack

Name: gi-dellav/zerostack review
Item: gi-dellav/zerostack
Rating: 70
Author: Hlido Editor

STEADY · 70/100

A lean Rust coding agent built for low memory and speed, with a genuinely clever parallel-worktree design — promising and fast, but earlier-stage than the incumbents.

Visit gi-dellav/zerostack →

Hlido Editor · 2026-07-21

Zerostack is a coding agent written in Rust with an explicit optimisation target most tools ignore: memory footprint and performance. It wraps OpenRouter (default), OpenAI, Anthropic, Gemini and Ollama, streams tokens, and reads/writes/edits/searches your codebase with user-controlled permissions on what it may touch. Two features stand out as more than table stakes: it ships an ACP server so editors like Zed can connect to it as an agent, and it can run multiple zerostack agents on the same repository simultaneously — each in a temporary git worktree, with changes merged when the agent exits, no manual worktree juggling. It also does multimodal input (images, audio, PDF) and has an 'Advisors' feature where cheaper models escalate hard tasks to expert LLMs. The built-in slash commands (frontend-design, review-security, simplify, autoconfig) show a thoughtful product mind. The honest framing: this is GPL-3.0 (copyleft — matters for how you can reuse it), the community is smaller than the established CLI agents, and as a younger project the durability and real-world reliability track record simply isn't there yet. It's a credible, well-designed entrant rather than a proven default.

Why STEADY

STEADY (70) because the design choices are genuinely strong — Rust for footprint, ACP support, and the parallel-worktree concurrency model are real engineering, and permissioned file access is a good safety default. Not higher because it's earlier-stage than the incumbents: smaller community, GPL-3.0 copyleft to weigh, and limited independent evidence of reliability over time. A tool to watch and try, not yet a safe default.

What it does well

Rust implementation optimised for low memory footprint and speed
Runs multiple agents on one repo in parallel via automatic temporary git worktrees
ACP server support — connects to editors like Zed as an agent
Permissioned file access: user controls what the agent may read/write
Multimodal input (image/audio/PDF) and an Advisors escalation path to stronger models

What it fails at

GPL-3.0 copyleft — a real reuse consideration for commercial/proprietary contexts
Smaller community and shorter track record than established CLI coding agents
Independent reliability evidence over time is limited (younger project)
No hosted UI — CLI/editor-oriented, assumes developer fluency

Best for

Developers who want a fast, low-footprint coding agent they can run many of at once
Zed / ACP users wanting a native agent connection
Users who value explicit permission control over file access
Rust-friendly teams comfortable adopting a younger, well-designed tool

Not recommended for

Teams that need GPL-3.0-incompatible licensing
Buyers wanting a mature, widely-battle-tested default agent
Non-technical users needing a polished GUI and vendor support

Compared to

aider footprint-and-concurrency
Aider is the mature, widely-adopted git-native pair programmer; zerostack is the leaner, faster Rust newcomer with a novel parallel-worktree model. Choose Aider for a proven default, zerostack for footprint/performance and concurrent agents.
gptme-gptme maturity-vs-performance
gptme is the older, broader Python terminal agent; zerostack is narrower and Rust-native with ACP and multi-worktree concurrency. gptme for breadth and maturity, zerostack for speed and parallelism.

Agent relevance

CLI Behavioral-testable

Agentic-Commerce Readiness 17/100 · CLOSED

Independent readiness for agent delegation & transaction. How it’s scored · check live

Good. Ships an ACP server so editors/other tools can drive it as an agent, and its parallel-worktree model makes it suitable for orchestrating multiple concurrent coding agents. CLI-first; no REST API or MCP surface documented yet.

Agent-friendly score: 6/10

What we saw

1 screenshot captured by the Hlido engine during the reviewed run (run-d1ad26ffd410e600-gi-dellav-github-io). Our own captures — not vendor marketing material.

gi-dellav/zerostack — run screenshot 1 (home.png) — home.png

Evidence

Lightweight Rust coding agent optimised for memory/performance — source (2026-07-21) verified
Wraps OpenRouter/OpenAI/Anthropic/Gemini/Ollama; ACP server support — source (2026-07-21) verified
Parallel agents via temporary git worktrees; permissioned file access — source (2026-07-21) verified
GPL-3.0 licensed — source (2026-07-21) verified

Public-surface checklist

✓ homepage_loads (required)
✓ primary_value_prop (required)
✗ cta_present (required)
✗ pricing_or_access
✗ evidence_or_demo

scorecard.json · registry · methodology

Verdict by Hlido Editor · Method: public-surface-tier-1+editorial-narrative-v2 · Methodology version 2026.07 · Next review due 2026-10-19

Embed this trust badge

Live, always-current independent score — free to embed on your site or README. No vendor pays for placement.

Markdown

[![Hlido trust score](https://hlido.eu/badge/gi-dellav-zerostack.svg)](https://hlido.eu/check/?agent=gi-dellav-zerostack)

HTML

<a href="https://hlido.eu/check/?agent=gi-dellav-zerostack"><img src="https://hlido.eu/badge/gi-dellav-zerostack.svg" alt="Hlido trust score"></a>