跳转至

Comparison

How does zhi compare to other terminal AI tools? This page provides a factual comparison across key dimensions.

Feature Matrix

Feature zhi aider Claude Code GitHub Copilot CLI Gemini CLI shell-gpt Qwen Code
Installation pip pip npm gh extension npm pip pip
Setup time ~30s (1 API key) ~2 min ~1 min GitHub login Google auth ~1 min ~1 min
Primary use Terminal AI + document processing AI pair programming Code agent Shell suggestions Code + terminal Shell AI Code agent
LLM provider Zhipu GLM (Chinese) OpenAI, Anthropic, etc. Anthropic Claude OpenAI / Anthropic / Google Google Gemini Various via OpenAI API Alibaba Qwen
File access Read + write + OCR (isolated output) Read + edit (in-place) Read + edit (in-place) None Read + edit None Read + edit
Skill system 15 YAML skills + custom None Slash commands None None None None
Cost model Dual-model (flash for skills, ~10% cost) Single model Single model + subscription Subscription Free tier + paid Single model Free
Chinese LLM Native (GLM-5 / GLM-4-flash) No No No No No Native (Qwen)
Output safety Isolated output dir, shell confirmation In-place file editing Sandbox + approval N/A In-place editing N/A In-place editing

Detailed Comparison

vs aider

aider is an AI pair programming tool focused on code editing within git repositories.

Aspect zhi aider
Focus Document processing + general terminal AI Code editing in git repos
File editing Writes to isolated zhi-output/ directory Edits files in-place, auto-commits
OCR Built-in (PDF, images via Zhipu API) No
Skills 15 reusable YAML workflows No equivalent
LLM Zhipu GLM (optimized for Chinese) Multi-provider (OpenAI, Anthropic, etc.)
Git integration Via shell tool Deep git integration

Choose zhi for document processing, Chinese language tasks, and reusable workflows. Choose aider for AI-assisted code editing with git integration.

vs Claude Code

Claude Code is Anthropic's official CLI agent powered by Claude models.

Aspect zhi Claude Code
Focus Document processing + Chinese LLM Code agent + general development
LLM Zhipu GLM Anthropic Claude
Cost Dual-model (~10% cost for skills via flash) Single model pricing
Skills 15 YAML + composable Slash commands (not composable)
File safety Isolated output directory Sandbox with approval prompts
Chinese Native GLM support English-focused

Choose zhi for Chinese document workflows and cost-efficient batch processing. Choose Claude Code for code-heavy development tasks with Claude's reasoning.

vs GitHub Copilot CLI

GitHub Copilot CLI suggests shell commands and explains terminal output.

Aspect zhi GitHub Copilot CLI
Focus Full agentic loop with tools Shell command suggestions
File access Read, write, OCR None
Autonomy Multi-turn agent with tool calls Single suggestion
Skills 15 reusable workflows None
Pricing Pay-per-token (Zhipu) GitHub subscription

Choose zhi for complex multi-step tasks. Choose Copilot CLI for quick shell command suggestions.

vs Gemini CLI

Gemini CLI is Google's terminal AI tool powered by Gemini models.

Aspect zhi Gemini CLI
Focus Document processing + Chinese LLM Code + general terminal
LLM Zhipu GLM Google Gemini
File editing Isolated output In-place
Skills 15 YAML + composable None
Pricing Pay-per-token Free tier available

vs shell-gpt

shell-gpt is a lightweight CLI for AI-powered shell commands.

Aspect zhi shell-gpt
Agent loop Multi-turn with tools Single-turn
File access Read + write + OCR None
Skills 15 reusable workflows None
Shell safety Blocked/warned/allowed tiers Basic

vs Qwen Code

Qwen Code is Alibaba's code agent powered by Qwen models.

Aspect zhi Qwen Code
Focus Document processing Code editing
Chinese LLM Zhipu GLM Alibaba Qwen
File editing Isolated output In-place
Skills 15 YAML + composable None
Pricing Pay-per-token Free

What Makes zhi Unique

Native Chinese LLM + Document Processing

zhi is the only terminal AI tool that combines a native Chinese LLM (Zhipu GLM) with a full document processing pipeline including OCR, table extraction, and format conversion. This makes it ideal for Chinese-language office workflows.

Dual-Model Architecture

zhi uses two models strategically:

  • GLM-5 (premium) for interactive chat -- intelligent, nuanced responses
  • GLM-4-flash (economy) for skill execution -- fast, cost-effective (~10% the cost)

This means batch operations like summarizing 50 documents cost a fraction of what a single-model approach would.

YAML Skill System

Skills are composable YAML configurations, not hardcoded features. You can:

  • Create custom skills without writing code
  • Compose skills together (e.g., contract-review uses analyze + compare + proofread)
  • Share skills as .yaml files

Safety by Design

zhi cannot delete your files — not because of a setting you might forget to enable, but because no delete capability exists in the architecture. There is no delete tool, no overwrite mode, and no way for the AI to remove or modify your original files. Safety is enforced in code, not in prompts — so it cannot be lost during context compression or ignored by the model.

This is a deliberate design choice: AI agents should never be able to cause irreversible damage to your data.

Output Isolation

Unlike tools that edit files in-place, zhi writes all output to an isolated zhi-output/ directory. Your original files are never modified, making it safe to experiment without risk.