Coding Agent
Published Updated
What is Coding Agent?
A coding agent is the tool that actually runs a prompt and writes the code, such as Claude Code, Codex CLI or Gemini CLI. In Orbit you register at least one and map each complexity tier to an agent and model.
An external CLI tool (Claude Code, Codex CLI, Gemini CLI) registered with Orbit to actually execute queued prompts. Configured per complexity tier.
Settings → Execution Tiers (also called "Coding Agent" in some surfaces) lets you register agent binaries — the path to your `claude`, `gemini`, or `codex` executables. You then map each complexity tier to a specific agent + model. Without at least one registered agent, no prompts can run. The Coding Agent settings include a Power Up UI helping users connect Claude Max/Pro subscriptions or API key configurations.
How it relates to Orbit
Orbit launches a registered coding harness for a queued prompt and displays its live session and transcript. See Agent Sessions & Harnesses for details.
When you would use it
You configure the Coding Agent during initial setup and revisit it when you want to change which tool runs your prompts.
Common operations
- Initial Orbit setup connecting your first agent
- Switching agents per tier as preferences shift
- Configuring Claude Authentication (subscription vs. API key)
Frequently asked questions
Which coding agents does Orbit support?
Claude Code, Codex and Gemini CLI, plus native CLIs for CodeWhale, Grok and OpenCode, and local or custom OpenAI-compatible endpoints. See Agent Sessions & Harnesses.
Do I need an API key to connect a coding agent?
No. Orbit can drive Claude Code through an existing Claude Max or Pro subscription, use Apollo as the default, or use your own key. See Connect a Coding Agent.
Can I see what the agent is doing while it runs?
Yes. Each run opens an Agent Session with its live turns, model, effort and context use.
Related terms
Where this is taught
No learning path uses this term yet. Browse the Learning Atlas for guided sequences through related ideas.