AI Environment

CLI Tools Overview & Comparison

The right agentic CLI depends on the research task, model access, and tolerance for setup complexity.

The best CLI agent is not the most powerful one, but the one that fits your research workflow.

Overview

The Research Memex workflow works with both GUI and CLI tools. GUI tools suit early exploration and teaching; CLI tools suit file-based, repeatable, auditable work. Most researchers should start with one of each: Cherry Studio for model comparison and knowledge bases, plus a CLI agent for project files, scripts, and long-running workflows.

Current CLI options

Cherry Studio

Visual multi-model workspace
Good first interface for API keys, knowledge bases, and MCP servers
Useful for teaching, comparison, and non-terminal researchers
Full guide: Cherry Studio Setup

Claude Code

Anthropic's terminal agent
Strong fit for deep project work, code, and research-writing infrastructure
Best when you want Claude as the main workspace model
Full guide: Claude Code Setup

Antigravity CLI

Google's current terminal-agent path for consumer and free Gemini CLI users
Good fit for Google-account workflows, SSH use, and Antigravity 2.0 continuity
Best for terminal-agent workflows, not general Gemini API access
Full guide: Antigravity CLI Setup

OpenCode

Multi-provider terminal UI
Good fit for comparing GPT, Claude, Gemini, DeepSeek, Kimi, GLM, Qwen, xAI, and local models
Best when model choice is the experiment
Full guide: OpenCode Setup

Comparison table

Feature	Cherry Studio	Claude Code	Antigravity CLI	OpenCode
Interface	GUI	CLI	CLI	TUI
Provider model	Many providers	Claude	Google Antigravity/Gemini paths	Many providers
Best role	Exploration and teaching	Primary project workspace	Google terminal-agent workflow	Provider comparison
Direct API keys	Yes	Usually no	No for normal sign-in flow	Yes
MCP support	Full	Full	`agy mcp add` or documented settings	Depends on host/provider config
Project memory	Knowledge bases and chats	Project files + Claude config	Project files + Antigravity settings/plugins	Project config
Cost shape	Provider API usage	Subscription/API, depending on plan	Google account/enterprise path	Provider API usage
Best for	Starting out	Deep work	Google-agent continuity	Model experimentation

Which one should you use?

If you are new to the workflow

Start with Cherry Studio. It makes providers, API keys, knowledge bases, and MCP servers visible. Add a CLI only after you have a real project folder and repeated tasks.

If you want a project workspace

Choose an interface that supports your project files and the models you want to work with. The Claude Code guide covers one route. The Codex access notes cover another. Kimi can also participate through its own workspace or a plugin, as described below.

Working with Kimi

Kimi is a regular collaborator in the author's reasoning, writing, coding, and sustained projects. Its different perspectives matter throughout the work, including how a problem is first framed.

Keep the model and its access routes distinct:

Kimi models: choose the model you want to work with; the model reference describes the family
Kimi Code: use the native workspace to work directly with the configured model. Follow the Kimi Code getting started guide (opens in a new tab) for setup.
Kimi Plugin: use the Claude Code or Codex integration within your existing workspace. Ask Kimi a question, invite a review, or delegate a task.

The plugin is one route into collaboration with Kimi. A review is one possible role. See a workflow that begins with Kimi for another.

If you used Gemini CLI

Move the terminal-agent workflow to Antigravity CLI. Keep direct Gemini API work with Google AI Studio, Cherry Studio, OpenCode, Vox, or another provider client.

If you want to compare models

Use OpenCode. It keeps the tool constant while you vary the model or provider. This approach works well for model-selection experiments.

Research phases

Research phase	Recommended surface	Reason
First setup	Cherry Studio	Lowest friction for provider keys and knowledge bases
Literature exploration	Cherry Studio or OpenCode	Easy model comparison and corpus experiments
File-based analysis	Claude Code or Antigravity CLI	Agents can work inside a project folder
Multi-model challenge	OpenCode or Vox MCP	Same prompt across different providers
Final writing and infrastructure	Claude Code	Strong project reasoning and repo operations

Terminal basics

If you are new to terminals, these commands are enough to begin:

cd ~/Documents/Research-Project
pwd
ls

Launch the CLI you installed:

claude
agy
opencode

Exit most terminal tools with exit, /exit, or Ctrl+D, depending on the tool.

Do not use the same authentication pattern for every tool.

Auth pattern	Used by	Notes
Provider API key	Cherry Studio, OpenCode, Vox	Use `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `GOOGLE_API_KEY`, `GEMINI_API_KEY`, etc. only when the client calls that provider directly.
Product sign-in	Claude Code, Antigravity CLI	The tool handles the session. Do not invent provider env vars unless the official docs say to.
Enterprise project binding	Antigravity CLI, some Claude/Google setups	Follow your organization's onboarding path.

@misc{docs-implementation-ai-environment-setup-cli-setup-guide-2026,
  author = {Xule Lin},
  title = {CLI Tools Overview \& Comparison},
  year = {2026},
  howpublished = {\url{https://research-memex.org/docs/implementation/ai-environment-setup/cli-setup-guide}},
  note = {ORCID: 0000-0001-7885-4194}
}

one renderingthe source remains