Default Model per CLI
- Desktop app
- All plans
What it does
An agent left on Default runs whatever its CLI picks: the default of your account type on Claude Code, Gemini 3.8 Flash on Antigravity. Default model lets you choose, once per CLI, the model every agent, ticket, team step and scheduled trigger without a model of its own launches on, on every project. Without a choice nothing changes: the CLI keeps its own default. Full page: https://agentsroom.dev/features/default-model
The model an agent starts on is read at launch, from the most specific to the most general: the model the agent, ticket or team step names, then the project's default model, then the Default model of Settings > AI providers, then the CLI's own default. A model switched off in Models available to agents is never launched: the first model still on starts instead.
Video (1 min 16): Default model, shown in the real app. In Settings, AI providers, Claude tab, the Default model is set to Sonnet 5.5; a new agent whose Model card is left on Default then starts on Sonnet 5.5, as the Claude Code banner and the header pill show.
Where to find it
- Global: Settings > AI providers > the CLI's tab > Default model, just under Models available to agents.
- Per project: Project settings > AI providers, pick the provider, then Default model under the list of models.
- The field only appears for a CLI that offers at least two models to choose from.
How to use it
- Open Settings > AI providers and select the CLI's tab (for example Claude).
- Under Default model, click a model chip. The choice is saved at once, there is nothing to confirm.
- Start a new agent of that CLI with its Model card on Default ("Provider's pick"): it launches on the model you chose.
To let the CLI pick again, choose Default in the same field. In the project field, Default means "follow the global setting".
Settings
defaultModels(global, synced with the account, Settings > AI providers > CLI tab > Default model): per provider, the model a launch left on Default uses. Empty by default (each CLI keeps its own default).defaultModels(project override, Project settings > AI providers > Default model): replaces the global default for this project's agents.- Related:
hiddenModels(Models available to agents), see Multi-Provider Support.
Agent tools (MCP)
settings_set: writesdefaultModels, for example{"antigravity": "gemini-4-argon-medium"}; withscope=project, the project override.settings_get: reads the current value.agents_spawn: an agent spawned without a model follows the default model too.
Providers
Every CLI whose model picker lists at least two models (Claude Code, Codex and Antigravity among them). The value is a model id of that CLI; a model id means nothing to another CLI, so each provider has its own entry.
Mobile
The phone has no screen to edit it. Agents and tickets started from the phone launch through the desktop, so they use the default model too.
Limits
- Applied at launch only: an agent already running keeps its model until it is restarted.
- A model pinned on the agent, the ticket or the team step always wins. The saved agent stays on Default, so clearing the setting gives it the CLI's choice back.
- The predefined roles of the Add agent window come with their own model (Frontend and Backend on Opus, for example): to make such a new agent follow the Default model, open its Model card and choose Default.
- A
/modeltyped inside the CLI itself is not affected.
Common questions
- I set a Default model but my new agent still starts on Opus. The role you picked in Add agent carries its own model. Open the Model card of the new agent and choose Default ("Provider's pick"): it then launches on your Default model.
- My Antigravity agents always start on Gemini 3.8 Flash. Can I change that for all of them? Yes: Settings > AI providers > Antigravity tab > Default model, pick the model (for example Gemini 4 Argon). Every Antigravity agent left on Default now starts on it.
- Can a project use another default model? Yes, in Project settings > AI providers > Default model. Choosing Default there follows the global setting again.
- Do running agents change model? No. The setting is read when an agent starts; restart the agent to apply it.
- Is there a video that shows it? Yes, 1 min 16 in the real app: https://youtu.be/Sb6RLHU9Z-s
Related
- Multi-Provider Support: the CLIs, their model pickers and Models available to agents.
- Quota Rules: rules that switch new launches to another model or provider.
- Agent Permissions (Autonomous or Manual): the other per-CLI setting of Settings > AI providers.