Claude Code

Anthropic’s coding CLI — via the Anthropic API, an internal gateway, or a cloud identity.

Claude Code (claude) is Anthropic’s agentic coding CLI. It’s the default agent and the only one that supports the full range of providers — so you can point it at the Anthropic API directly, at an internal gateway, or at a cloud model backend that authenticates with workload identity.

Providers

Set the provider once per org in Admin → Agent LLM:

Provider	How it authenticates	Needs a stored key?
Anthropic API	Direct API with your Anthropic key	Yes
Internal gateway	Anthropic-compatible proxy at your endpoint	Yes (endpoint + key)
AWS Bedrock	IRSA / workload identity (region from deployment)	No
Google Vertex AI	Workload identity (project/location from deployment)	No
Azure AI	Managed identity (resource from deployment)	No

Only Anthropic and Internal gateway need a stored API key. Bedrock, Vertex, and Azure authenticate through the cloud’s own identity — no key is stored, and the agent process never sees cloud credentials.

Set it up

1. Open Admin → Agent LLM

   Org-admin only.
2. Pick the provider

   Choose Anthropic API, Internal gateway, Bedrock, Vertex, or Azure.

3. Set the model

   A free-text field with suggestions (e.g. claude-opus-4-8, claude-sonnet-5). It’s free-text because Bedrock/Vertex take inference-profile IDs or ARNs.

4. Add the API key (Anthropic / gateway only)

   The key is write-only — you set it, and the UI only ever shows whether one is configured, never the value.

Using it in a task

Pick Claude Code in the New Task Agent dropdown. If your org has more than one Claude model allowed, a Model dropdown lets you choose per task — run one task on Opus and another on Haiku, all on the same configuration.

Claude Code also powers the non-code experiences (Cowork assistant and Design), which are pinned to Claude.

Using a gateway? Make sure the endpoint is the origin (no trailing /v1) — see Internal gateway for why.