Agents & models overview

Choose an agentic coding CLI per task, and configure the LLM backend once for your org.

Two independent choices shape how a task runs:

Agent — which CLI runs

The agentic coding CLI that runs in the workspace: Claude Code, Codex CLI, Gemini CLI, or Kiro CLI. Chosen per task.

Provider — which LLM backend

For Claude Code, the backend + auth path: Anthropic API, an internal gateway, AWS Bedrock, Google Vertex, or Azure. Configured once per org in Admin.

The agents

Agent	Credential	Where the key comes from
Claude Code (claude)	Anthropic API key	Org LLM key (or cloud identity / gateway)
Codex CLI (codex)	OpenAI API key	Org “OpenAI key (Codex)“
Gemini CLI (gemini)	Google API key	Org “Google key (Gemini)“
Kiro CLI (kiro)	None (device-flow login)	Logged in once, then reused

There’s a reserved custom agent for a future bring-your-own CLI. It isn’t selectable and can’t run yet.

Where you choose

• Per task — the New Task form shows an Agent dropdown (limited to the CLIs your org allows) and, when your org has more than one model configured, a Model dropdown. See Create a task.
• Per org — an admin sets the provider, model(s), keys, and policy under Admin → Agent LLM. Until an LLM is configured, tasks can’t be dispatched.

What to read next

Claude Code

Anthropic, gateway, or cloud identity.

Codex CLI

Runs on an OpenAI key.

Gemini CLI

Runs on a Google key.

Kiro CLI

Device-flow login, once.

LLM configuration

Provider, model, and keys.

Internal gateway

Route everything through one proxy.

Multiple models

Offer several models per org.