Tune your settings

Settings holds what’s yours, not your workspace’s. Open the account menu and choose Settings. Under Your sessions, these tabs shape how the sessions you start behave: Defaults, Instructions, Memory, Plugins, Hooks, Agents, and Integrations.

Your instructions, memory, plugins, hooks, and agents load into each session you start. On New, the Session row under More has a chip that says what the session will run with. On a phone, the same line is in the + drawer. Open it and choose Start clean to leave them behind for that one session.

Defaults

Defaults is for choices you’d otherwise make each time you start a session. A project can override them, and you can still change them in any session. They’re grouped like this:

• Sessions. Permission mode and When it needs an answer set how your new sessions run, and Remember my last mode starts them in the mode you last switched to. Session names tells the agent how to name a session you don’t name yourself. Overview sections chooses which Overview sections start open, and Overview order sets their order. See Autonomy for what the modes mean.
• Models. The model and thinking depth your new sessions start on, a row for each phase, Remember my last pick, and Use the org’s picks. See Choose a model and how hard it thinks.
• Slack. Where the agent talks with you in Slack, whether to Start every session in Slack, and whether everyone who posts in a thread joins the session. See Slack.
• Interface. Your Default group, which pages that show a group start with. How the agent talks to you is Automatic, Plain language, or Technical. Other rows cover saving a draft brief when you click away, commenting on plan text, opening Artifacts on your own, and suggesting groups from the people you work with.

Instructions

Instructions is a document you write once, in Markdown, about how you like to work. For example, “Lead with the failing test, then the fix.” Evergreen loads it into each session you start, next to the repository’s own instructions.

Where the two conflict, the repository’s instructions win, and so do Evergreen’s rules. Yours say how you like to work, not how a repo works. Choose How this loads on the tab for a short explanation.

Use Instructions for what you want stated. Use Memory for what the agent works out on its own.

Memory

The agent saves notes as it works with you, such as a preference or a pitfall, so the next session doesn’t start from nothing. Memory shows those notes as the agent wrote them.

• Notes About you apply to all your sessions.
• Notes about your work stay with the repository they came from. Each repository has its own group.
• Choose Edit to change a note, Forget this to remove one, or Forget everything here to clear a whole group.
• Use memory in my sessions turns memory on or off. When it’s off, sessions start without notes and nothing new is saved. Your existing notes stay until you delete them.

A project keeps its own memory too, which everyone on the project can read. See Group sessions into a project.

Plugins

A plugin adds skills, commands, agents, and hooks to your sessions, and loads the same way it would on your laptop.

1. Choose Browse plugins to search the stores you can reach.
2. Choose Install on one.
3. It loads into each session you start.

Each installed plugin has a switch, and a menu to check for updates, pin a version, view its source, or remove it. Marketplaces lists where plugins come from. You can add your own with Add from a repository. An Owner or Admin can decide which plugins are available, installed for everyone, required, or hidden for the whole workspace. The Variables card holds values a plugin’s servers ask for, such as an API key. Yours are stored encrypted.

Hooks

A hook runs a command of your own at a set point in a session. It might run before a tool call, after an edit, or when the agent stops. Your hooks run in each session you start.

• Choose Add hook to start from a recipe or write your own. Import reads hooks from a settings.json.
• Each hook has a switch, and its menu lets you edit, duplicate, copy it as JSON, or delete it.
• A hook that runs before a tool can refuse the call or ask you first. It can’t approve a call on its own.
• Hooks from your installed plugins show under From your plugins. Switch the plugin on or off to turn them on or off.

How hooks run explains how hooks in a session differ from hooks on your laptop.

Agents

Agents are subagents that the agent can hand a task to in the Claude Code sessions you start. You might define a code reviewer or a test writer.

Choose Add agent to write one, or Import to paste an agent file from .claude/agents. Each agent has a name, a description that tells the agent when to use it, the tools it may use, a model, and a prompt. The model is Same as session, Sonnet, Opus, Haiku, or Fable. An agent from a plugin appears under From your plugins and turns on and off with its plugin.

Integrations

Integrations is where you connect services such as GitHub, Slack, or Jira so the agent can act in them as you, with your permissions. The agent’s actions are attributed to you, and disconnecting a service deletes its stored tokens. A Ready to connect list shows the services you can connect now, and Browse the marketplace helps you find more.

See Integrations for what each one does and how to connect it.

Other tabs

Under Account, Settings also has Profile, Preferences, Notifications, and Devices, and Workspaces when you belong to several workspaces, have one you can join, or can start another. Notifications is where you choose what reaches your phone and Slack. See Use Evergreen on your phone. The Help group at the foot opens Keyboard shortcuts, What’s new, Documentation, and Contact support.