Zeltro › AI coding agents that work with Zeltro, from Claude Code to Aider

Every AI coding agent Zeltro works with

Zeltro works with seven AI coding agents: Claude Code, Codex, Antigravity, Qwen Code, Aider, OpenCode and Hermes Agent. How each one signs in, which can run a free local model, and how to pick one.

Zeltro doesn't sell AI. It runs the coding agent you choose, on your own computer, inside a project that is already set up: the right runtime, a database, a working .env. The agent changes the files, and you watch the result in the app.

There are seven agents to choose from, plus one that is on its way out. Zeltro installs whichever you pick, so there is nothing to set up by hand, and you can keep several profiles and give each project its own.

• Claude Code
• Codex
• Antigravity
• Qwen Code
• Aider
• OpenCode
• Hermes Agent

At a glance#

All of them except Antigravity, which talks only to Google, can also reach hundreds of models through OpenRouter.

The agents#

Anthropic

Claude Code

The agent Zeltro suggests first. It signs in with your Claude Pro or Max plan, so a subscription you already have does the work, or with an Anthropic API key. Zeltro handles the sign-in for you.

• Commandclaude
• Local modelsThrough an Anthropic-compatible endpoint
• Tells the app when it's doneYes

claude.com

How Zeltro runs it

OpenAI

Codex

OpenAI's coding agent. Sign in with your ChatGPT plan or an OpenAI API key, and Zeltro does the sign-in for you. It also runs local models, with one-click presets for Ollama and LM Studio.

• Commandcodex
• Local modelsYes
• Tells the app when it's doneYes

developers.openai.com

How Zeltro runs it

Google

Antigravity CLI

Google's terminal agent, and the successor to Gemini CLI. Sign in with your Google account by running agy once in a terminal, or paste a Gemini API key into the profile. New in Zeltro 1.0.0-beta.34.

• Commandagy
• Local modelsNo
• Tells the app when it's doneNot yet

antigravity.google

How Zeltro runs it

Qwen team, Alibaba

Qwen Code

Built around the Qwen Coder models, and a good low-cost choice. It needs an API key or an endpoint and a model name. The cheapest path we have tested end to end is Qwen Code with Qwen3 Coder on OpenRouter.

• Commandqwen
• Local modelsYes
• Tells the app when it's doneNot yet

github.com/QwenLM

How Zeltro runs it

Open source

Aider

The veteran open-source pair programmer. It has no subscription sign-in: give it an API key and a model name, or point it at Ollama or LM Studio and it runs entirely on your own hardware.

• Commandaider
• Local modelsYes
• Tells the app when it's doneYes

aider.chat

How Zeltro runs it

Open source

OpenCode

An open-source agent that works with many providers. Sign in with a ChatGPT plan, an API key or an endpoint by running opencode auth login once. Models are written as provider/model, such as openai/gpt-5.4-mini.

• Commandopencode
• Local modelsYes
• Tells the app when it's doneNot yet

opencode.ai

How Zeltro runs it

Nous Research

Hermes Agent

Nous Research's agent. Sign in with a Nous Portal or ChatGPT plan, an API key or an endpoint by running hermes setup once. Like OpenCode, it takes models as provider/model.

• Commandhermes
• Local modelsYes
• Tells the app when it's doneNot yet

nousresearch.com

How Zeltro runs it

Google · on its way out

Gemini CLI

Still works in Zeltro with a paid Gemini API key, but new profiles can't pick it. Google stopped serving it to personal accounts on 18 June 2026. Switch to Antigravity, below.

• Commandgemini
• Local modelsNo
• StatusDeprecated

github.com/google-gemini

"Tells the app when it's done" means the agent reports back when it finishes a turn or needs your answer, so the Zeltro app can show it. The others work just as well; you just watch the terminal for that.

Which one should I pick?#

Not sure? Start with the one whose plan you already pay for. You can add more profiles later and give a project a different one.

Local and low-cost models#

Five of the agents can run a model on your own computer, at no cost beyond your hardware. The Zeltro app has one-click presets for these:

• Ollama
• LM Studio
• OpenRouter

Those five are Codex, Qwen Code, Aider, OpenCode and Hermes Agent. Claude Code can use other models only through an Anthropic-compatible endpoint, and Antigravity can't use them at all.

A small model on a laptop is much weaker at building a whole app than a frontier model. OpenRouter sits in between: open models such as Qwen3 Coder, run in the cloud, for a fraction of the price. See cheap and local models for the settings.

Moving from Gemini CLI to Antigravity#

Google stopped serving Gemini CLI to personal Google accounts (free, AI Pro and AI Ultra) on 18 June 2026. A paid Gemini API key still works with it. If you picked Gemini CLI in an older Zeltro:

1. Update Zeltro

   Restart the Zeltro app, which updates it to 1.0.0-beta.34 or later.

2. Choose Antigravity CLI

   In Settings → AI Profiles, edit your profile or add a new one, and choose Antigravity CLI. Zeltro installs it.

3. Sign in

   Run agy once in a terminal and sign in with Google, or paste a Gemini API key into the profile. The first start asks for a color scheme and shows Google's terms and data-use screen. Read it before you accept: the box for sharing data with Google is ticked by default, and Zeltro leaves that choice to you.

Setting it from the terminal#

The app's Settings → AI Profiles does all of this, but the same settings are one command away:

zeltro ai-set --agent claude
zeltro ai-set --agent agy
zeltro ai-set --agent qwen --model qwen/qwen3-coder-next --api-base https://openrouter.ai/api/v1 --api-key sk-or-...
zeltro ai-set --agent aider --model openai/llama3.1 --api-base http://localhost:11434/v1 --api-key ollama

By default, the AI in each project works without stopping to ask before each change: the Works without asking switch in the project's panel is on, and you can switch it off project by project to approve each step in the agent's terminal. Zeltro passes that choice to the agent for each run rather than changing the agent's own settings. To change an agent's own setting, outside Zeltro too, run zeltro ai-unattended --agent <name> on. Working with AI has every option.

Questions

Checked 2026-10-24. Other products change often — if something here is out of date, tell us and it gets fixed.