Which is a Higher Coding Agent?

Open-source CLI instruments have created lots of pleasure within the developer and AI communities. With the discharge of Claude’s Codex CLI, OpenAI’s Codex CLI, and, extra not too long ago, Google’s Gemini CLI, there’s a sudden enhance in curiosity in CLIs typically. By incorporating AI assist into routine duties, these instruments are altering how builders use their terminals.

Builders can enhance code comprehension, pace up debugging, and automate tough coding duties by utilising these CLIs. Instruments like Codex CLI assist customers write, analyse, and enhance code extra effectively by bringing sturdy AI-driven options like code era, refactoring, and inline documentation instantly into the command line. The market has grow to be much more thrilling with the launch of Gemini CLI, a light-weight, open-source substitute from Google that’s gaining reputation rapidly.

Let’s now take a more in-depth have a look at how Gemini CLI and Codex CLI fare when put to the take a look at.

What are Gemini CLI and Codex CLI?

Gemini CLI: An open-source AI agent by Google that integrates Gemini fashions (like Gemini 2.5 Professional) into the terminal for builders.

Codex CLI: An open-source command-line device forked from OpenAI’s authentic Codex CLI, now supporting a number of AI suppliers, together with Gemini 2.5 Professional. It permits builders to work together with massive language fashions within the terminal for duties like code era, refactoring, and script rationalization.

Should you’re on the lookout for detailed walkthroughs, take a look at our guides on every CLI device:

Set up

Earlier than continuing with the set up, be certain that the system necessities are glad.

System Necessities

NodeJS is a standard prerequisite for each the CLIs. You’ll be able to set up Node.js by following the directions on its official web page: https://nodejs.org/

Set up Steps for Codex CLI

1. Test Node.js model:

node --version

2. Guarantee it’s NodeJS model 22 or greater.
3. Set up Codex CLI globally:

npm set up -g @openai/codex

4. Or utilizing pnpm:

pnpm add -g @openai/codex

5. Set your OpenAI API key (if inside your venture listing’s .env file accommodates the OPEN AI API Key, then we don’t want to do that process):

export OPENAI_API_KEY="your-openai-api-key"

6. Add this line to your shell config file (e.g., ~/.bashrc, ~/.zshrc) for persistence.
7. Run Codex CLI:

codex

8. Or cross a immediate instantly:

codex "clarify this codebase to me"

Set up Steps for Gemini CLI

1. Test Node.js model:

node --version

Guarantee it’s 18 or greater.

2. Set up Gemini CLI globally (really helpful):

npm set up -g @google/gemini-cli

3. Alternatively, run instantly with out set up:

npx https://github.com/google-gemini/gemini-cli

4. Authenticate:
   1. Default: Sign up along with your Google account when prompted (browser window will open).
   2. API Key (for superior use):
      1. Get your API key from Google AI Studio.
      2. Set it in your surroundings utilizing the next entry:
         export GEMINI_API_KEY=”your-api-key”
5. Run Gemini CLI:

gemini

6. Or cross a immediate instantly:

gemini "Hi there, are you able to assist me with coding?"

Issues to Maintain in Thoughts

• For each instruments, Node.js and npm (Node Package deal Supervisor) have to be put in and obtainable in your system PATH.
• For Codex CLI, Node.js 22+ is required, whereas Gemini CLI works with Node.js 18+.
• Each require authentication through API keys or account login for full performance.
• On Home windows, Codex CLI is finest run through WSL2 for full compatibility.
• Keep in mind, while you wish to use your CLI for a sure venture, be sure you have the present working listing set to the folder path the place your venture exists. 

Operating our CLI Situations

Let’s proceed to match the efficiency between the two fashions with 3 duties

• Constructing a working mannequin on an area host
• Studying a posh codebase and explaining the code
• Error debugging

Let’s first strive it out with Codex CLI:

Codex CLI

If all the necessities have been glad, then we are able to begin with the checks.

Activity 1 – Explaining our Codebase

We are able to see that Codex can present an in depth abstract of our code base, which was, to be sincere, surprisingly good. Moreover, Codex CLI briefly asks us to grant entry to particular code segments, which we are able to grant or reject based mostly on our judgment. It offers a good report after taking an inexpensive period of time to replicate.

The report’s high-level goal, repository format, key elements, front-end web sites, and property are all included, together with a abstract. Regardless that I had given the principle recordsdata quite ambiguous names, I used to be nonetheless shocked to seek out that it might decode the entire earlier iterations of the code recordsdata in chronological order.

Activity 2 – Error Debugging

Since we enable the Codex CLI entry to our complete database, whether or not or not it’s personal or protected, it would ask for our authorization earlier than studying any code snippets. If we allow it, it would entry our snippet appropriately. 

Total, it was capable of recognise each minute mistake and sometimes provided fixes for some careless errors within the code, eradicating pointless parts that had an impression on a few of the latency barely.

Activity 3 – Additions to Codebase

This concerned understanding how every module was outlined throughout varied directories and bridging the data between every dependency to make sure that a brand new listing was created to construct the new.py recordsdata for the additions we had requested after we requested entry.

Let’s now try to look at how the Gemini CLI features.

Gemini CLI

I preferred the Dracula theme.

Not like Codex CLI, the place we solely have to export our OPENAI_API_KEY, we have now three alternative ways to log in right here: utilizing our Gmail account, copying the GEMINI_API_KEY from AIStudio, or utilizing Vertex AI credentials.

Activity 1 – Clarify our Codebase

The primary distinction I noticed was that, in contrast to Codex CLI, which asks for entry to particular directories, Gemini skips this step and as an alternative scans our complete database. However, in my expertise, it does provide a good overview, however it’s under no circumstances nearly as good as Codex by way of output.

Activity 2 – Error Debugging

In my view, Gemini’s response was considerably higher defined than Codex’s, because it was capable of determine the sorts of safety dangers, clarify them, and supply file names as references for additional investigation and steered adjustments. Moreover, providing a radical technique to handle every of those errors would allow the code to be modified with out impairing its performance if the mandatory directions had been adopted.

Activity 3 – Additions to Codebase

Gemini CLI wasn’t working for this workload, so no output was produced.

Gemini CLI vs Codex CLI: Last Verdict

Additionally Learn: 10 Methods College students Can Use Cursor AI for Free

Conclusion

When evaluating the options of those two CLIs, I discovered that Codex provides extra performance than Gemini CLI. In the mean time, each the CLIs are open-sourced and can ultimately have higher performance in future releases. 

Since a GPT mannequin analyses our code, Codex additionally provides larger privateness by requesting entry to our codebases and modules in a method that minimizes safety danger. Nevertheless, this doesn’t seem like a case of the Gemini CLI requesting entry usually; as an alternative, it asks when it needs to run the applying code. Moreover, we must always do not forget that the context on the suitable aspect of the terminal can also be tracked by Gemini CLI. 

When mixed with extra clever vibe coding IDEs like Cursor, Windsurf, and others, such CLIs would undoubtedly open up new potentialities for vibe coding workflow and software operation. Nevertheless, safety and privateness vulnerabilities should exist, so I might advise being cautious about what and the place in our codebase we grant these CLIs entry.  

Continuously Requested Questions

GenAI Intern @ Analytics Vidhya | Last Yr @ VIT Chennai
Keen about AI and machine studying, I am desirous to dive into roles as an AI/ML Engineer or Information Scientist the place I could make an actual impression. With a knack for fast studying and a love for teamwork, I am excited to deliver modern options and cutting-edge developments to the desk. My curiosity drives me to discover AI throughout varied fields and take the initiative to delve into knowledge engineering, making certain I keep forward and ship impactful tasks.