Default Reasoning Effort
- Desktop app
- All plans
What it does
Reasoning effort is how hard the model thinks before it answers: more effort usually gives better plans on hard problems, and costs more time and more of your quota on every turn. Each CLI exposes its own lever, under its own name. Setting it agent by agent works until you run twenty agents across five projects, so AgentsRoom keeps one default per CLI: Claude Code on Very high, Codex on High, Kimi Code on Low to keep costs down. Every agent of that CLI that pins no effort of its own launches at that level.
The level is never written on the agents. Change the default and every agent that relies on it starts at the new level on its next launch.
Video (1 min 19): One reasoning effort per CLI, shown in the real app. The Add agent window shows Default with nothing inherited, Claude Code is set to Very high in the settings, a second CLI offers only the levels it accepts, and the next agent boots on the inherited level without anything set on it.
Where to find it
- Settings > AI providers > the CLI's tab > Default reasoning effort, under Agent permissions and Default context window. The quickest way in is the AI providers chip of the app's footer, right of the Settings one, which opens the settings on that page. The field is a slider whose steps are the levels that CLI accepts, plus a first step named Default, which leaves the CLI on its own setting. Right of that label, a small help icon with a play badge ("Watch the video tutorial on YouTube") opens the video of the feature in your browser. It is on the global page only, not in the project settings.
- Project settings > AI providers: one row per CLI with a lever. Default follows the global setting; a level picked there applies to this project's agents only.
- Per agent: the REASONING card of the Add agent and Edit agent windows, between MODEL and WORKSPACE. It reads Default, with the inherited level written under it (empty when no default is set anywhere).
- An agent's header and its right-click menu show the level the session
actually runs on, and the CLI writes it in its own banner when it starts
(
Opus 5 (1M context) with xhigh effort, andxhigh · /effortin its status line for Claude Code).
How to use it
- Open Settings > AI providers and pick the CLI's tab, for example Claude Code or Codex.
- Drag or click Default reasoning effort. Click a level's name under the slider to jump straight to it. The steps are named Low, Medium, High, Very high, Max and Ultra, and a tab only shows the ones its CLI accepts: Claude Code has five, Kimi Code has three.
- Override it for one project in Project settings > AI providers if that project needs a different level.
- Start or restart your agents: the effort is applied when an agent starts, so a running agent picks up the new level on its next launch.
Which level wins, first one found: the agent's own effort (a backlog ticket, a team step, a scheduled task or a temporary launch override counts as one), then the project default, then the global default, then the CLI's own default.
Settings
reasoningEffortDefaults(global, Settings > AI providers): one level per provider id, amonglow,medium,high,xhigh,maxandultra. A provider that is missing keeps the CLI's own default.reasoningEffortDefaults(project override, Project settings > AI providers): replaces the global level for this project's agents.reasoningEffort(per agent): the level pinned on one agent, ornullfor the default.botAgentReasoningEffort(global, Settings > Bot): the level of the Bot agent.
Agent tools (MCP)
settings_set: writesreasoningEffortDefaults, for the account or for one project, so an assistant agent can tune the defaults for you.agents_save: pinsreasoningEfforton a saved agent.agents_spawn:reasoningEffortsets the thinking level of an agent an agent creates (since 2026-10-28). Without it the new agent starts on the CLI's own default, however precisely the model was named; the answer names the provider, model, thinking level and account it actually runs on.teams_save:reasoningEfforton a team step.
Providers
Nine CLIs expose the lever, with the level each one accepts and the way AgentsRoom passes it:
| CLI | Levels | How the level reaches the CLI |
|---|---|---|
| Claude Code | low, medium, high, xhigh, max | claude --effort <level> |
| Codex | low, medium, high, xhigh, max (ultra on the models that offer it) | codex -c model_reasoning_effort=<level> |
| Grok Build | low, medium, high, xhigh, max | grok --reasoning-effort <level> |
| Antigravity CLI | low, medium, high, max (none on a model whose name already ends with its level, such as gemini-4-carbon-medium: that level is used) | --effort <level> |
| Kimi Code | low, high, max (none on the HighSpeed model) | KIMI_MODEL_THINKING_EFFORT=<level> |
| Command Code | low, medium, high, xhigh, max | --effort <level> |
| Muse Code | low, medium, high, xhigh, max, ultra | muse --reasoning-effort <level> |
| oh-my-pi | low, medium, high, xhigh, max | --thinking <level> |
| Pi | low, medium, high, xhigh, max | --thinking <level> |
The other CLIs AgentsRoom runs (GitHub Copilot CLI, Cursor, OpenCode, Aider, Mistral Vibe, Amp, Freebuff and Devin) have no effort lever here, so no default to set: the field does not show for them.
Mobile
No AI providers settings screen on the phone, so the defaults are set on the computer. Agents started from the phone still inherit them, because they run on the computer.
Limits
- The default is read when an agent starts: changing it does not touch an agent that is already running.
- A level the launched model does not offer is dropped, not moved to the nearest one: the agent starts on the CLI's own default.
- An
--effortyou typed yourself in a CLI's own options wins; AgentsRoom does not add a second one. - Effort is the CLI's own lever, not a token budget AgentsRoom enforces. What a level costs in time and quota depends on the CLI, the model and the task.
Common questions
- How do I set it for every Claude Code agent at once? Settings > AI providers > Claude Code > Default reasoning effort. Every Claude Code agent with no effort of its own then launches at that level.
- Which comes first, the agent, the project or the global default? The agent's own effort, then the project default, then the global default, then the CLI's own default. The first level found wins.
- Does changing the default restart my running agents? No. A running agent keeps its level until its next launch, and then picks up the new default if it pins none.
- Which CLIs support it? The nine with an effort lever: Claude Code, Codex, Grok Build, Antigravity CLI, Kimi Code, Command Code, Muse Code, oh-my-pi and Pi.
- Can an agent choose the thinking level of an agent it creates? Yes, since 2026-10-28: it passes
reasoningEfforttoagents_spawnalongside the provider, the model and the account, and reads back what the new agent actually runs on. - Is there a video that shows it? Yes, about a minute and twenty seconds, filmed in the real app: https://youtu.be/kpw42mfKgSU. Since 2026-10-28 the app links to it: the help icon with a play badge next to Default reasoning effort in Settings > AI providers.
- How do I check an agent really started on it? Open the agent and read the banner its CLI prints on the first line: it names the model and the effort it launched with.
- Does a lower effort save quota? Usually yes, since fewer thinking tokens are spent per turn, but the saving is the CLI's, not a cap AgentsRoom applies.
Related
- Multi-Provider Support: the CLIs and models the effort applies to.
- Adaptive Mode: picks the model for a task; the effort picks how hard that model thinks.
- Agent Delegation: an agent creating another one and pinning its engine.
- Claude Code Token Usage: the meter that shows what a level costs.
- Quota Rules: rules that lower the model or the effort when a quota runs low.