Overview
Default agent, model, and mode settings.
Settings → Agents is where you configure how each coding agent authenticates, which models it can use, and what a new chat starts with. Every supported agent (Claude Code, Codex, OpenCode, Grok, and others you've installed) gets its own page here.
Screenshot: Settings > Agents > Overview
The Agents overview page, listing every installed agent with a status badge and a chevron into its own settings page.
Overview
The landing page for the Agents scope. It lists every installed agent with its readiness status and a refresh action that reconciles local installs against the catalog. Selecting an agent opens its own page.
Per-agent pages: authentication and models
Each agent's page has a Cloud / Local switch, since your cloud sandboxes and your local machine can authenticate the same agent differently, and two tabs:
Screenshot: Settings > Agents > Claude Code (Authentication tab, Local)
A per-agent settings page with the Cloud/Local surface switch, the Authentication tab active, showing the Proliferate gateway toggle, API key rows, and a native sign-in action.
Authentication lets you choose how the agent reaches a model, per surface:
- Proliferate gateway. Route the agent's model calls through Proliferate's managed LLM gateway instead of a provider account directly. Available only when an admin has turned it on for the organization; see Allowed harnesses.
- API key. Supply your own provider API key. Some agents accept more than one provider (for example a fallback key), each as its own row.
- Native sign-in. Sign in to the agent's own account directly (Local surface only), the same as running the CLI on your own machine.
All models shows the agent's full model catalog with visibility toggles, so you can trim the composer's model picker down to the models you actually use, and a refresh action for agents that probe their catalog live.
Screenshot: Settings > Agents > Claude Code (All models tab)
The All models tab for a per-agent page, listing the agent's model catalog with per-model visibility toggles and a refresh action.
API keys
A single store of provider API keys, reusable across any agent's authentication rows instead of re-entering the same key per agent.
Screenshot: Settings > Agents > API keys
The API keys page, showing a list of stored keys with a redacted value hint and a form to add a new key by title and value.
Defaults
Sets the default agent for new chats, plus per-agent default model and launch controls (permissions mode, reasoning, reasoning effort, fast mode) where the agent supports them. See Defaults for the full breakdown.
These are personal settings. An org admin can still restrict which agents and authentication routes are allowed at all, independent of what you configure here.