Settings

Every panel in the Settings dialog, what each control does, and what your admin may have locked.

Open Settings from the title bar or the account menu. The dialog has six panels; this page covers each one.

If your IT admin pushed a managed policy, some fields show a lock icon — readable but not editable. Each section below calls out the common locks. For admins, see Reference → Policy fields.

General and About

The first panel. App-wide preferences plus version info and update controls.

Control	Default	What it does
Theme	Dark	Dark / Light / System. Applies across every agent.
Default model	Sonnet	Which model new chats use. Overridable per chat and per agent.
Minimize to tray	On	Window close keeps EasyCC running in the menu bar / system tray so scheduled tasks fire. Off stops scheduled tasks at window close.
Start on login	Off	Launch EasyCC at OS login. Pair with Minimize to tray for always-available.
Show hidden files	Off	Show `.easycc/`, `.claude/`, and other dot-folders in the file tree.
Hide thinking blocks / tool-call details / Compact mode	Off	Per-chat overrides under chat … menu; these are the global defaults.

Update controls:

Control	Default	What it does
Release channel	Stable	Stable (weeks), Beta (weekly), Alpha (multiple per week — unstable).
Defer updates	12 hours	How long the "Update available" banner can be postponed.
Auto-install on close	Off	Install pending updates when you quit EasyCC.
Auto-install on launch	Off	Install pending updates on next launch.

Common admin locks: minimize-to-tray, show-hidden-files, release channel, update endpoint, deferral caps.

Memory and learning

Controls for how Claude learns from you.

Control	Default	What it does
Let Claude learn from our chats	On	Weekly background review of recent chats; updates the preference memory. Off keeps existing memory on disk but stops new updates.
See what Claude picked up	—	Open the plain-English summary of learned preferences.
Save / Restore learned memory	—	Export `.easycc-memory.json` to back up or move between machines.
Forget this agent's learning	—	Wipe per-agent memory. Hand-written `CLAUDE.md` rules stay.
Forget shared layer	—	Wipe cross-agent preferences only.

Common admin lock: learning toggle locked off in regulated orgs (existing per-agent memory stays on disk but unused).

Safety and permissions

Two stacked sections. The safety scanner protects against prompt-injection in tool results; permissions control what Claude can do without asking.

Safety scanner

Control	Default	What it does
Cleanup	Always on	Rewrites invisible Unicode and wraps content in a clear marker. Free and instant.
Prompt injection detection	Off	On-device classifier; ~260 MB one-time download.
Sensitivity	Balanced	Relaxed / Balanced / Sensitive — false-positive vs recall tradeoff.
Double-check with Claude	Off	When flagged (small per-flag cost) / Always for web & connectors (more cost).
Block suspicious results	Off	Replace flagged content with a placeholder before Claude sees it. Only blocks when both detector and double-check agree.
Custom model source	(default — HuggingFace)	Pin to a mirror for air-gapped or proxy-restricted environments.

Permissions

Control	Default	What it does
Default permission mode	Ask	Applies to new chats. Ask / Auto-accept edits / Plan mode. Per-chat override stays.
Permission rules	—	Table of Allow / Ask / Deny patterns by file glob, command pattern, or URL. Most-specific rule wins.

Common admin locks: every safety field, plus specific deny rules and the default permission mode.

Integrations and connectors

Two stacked sections.

Integrations (MCP servers)

Three scopes:

Scope	Where it lives	Visible in
Project	`<agent>/.claude/mcp-servers.json`	This agent only
Personal	`~/.claude/mcp-servers.json`	Every agent
EasyCC app	App config	Every agent

Add a server from the Catalog (pre-vetted, one-click) or Custom (URL or stdio local server). OAuth happens in a browser; the credential lands in the OS keychain.

Toggle a server off to silence it without losing config; remove to delete entirely.

Cloud connectors like Firecrawl, Gmail, Calendar, monday.com. Their config lives on Claude.ai; this panel shows what's enabled and lets you toggle them. Auth happens through Claude.ai; revoke from the Claude.ai dashboard.

A slow connector is the most common cause of "chat is slow" — disable all, then re-enable one-by-one to find the offender.

Skills and library

Browse the 70 starter skills (Writing, Business, CFO, Marketing, Productivity, Learning, Creative, Engineering) and manage what's installed.

Action	What
Install in this agent	Writes to `<agent>/.claude/skills/<name>/SKILL.md`
Install for all agents	Writes to `~/.claude/skills/<name>/SKILL.md` instead
Edit	Change name, display name, description, tags, prompt, model override
Uninstall	Removes the folder

Installed skills show source (bundled / personal / per-agent), status badges (New, Updated, Failing, Suggested by Claude), and a Show in /commands toggle to hide a skill from chat autocomplete without uninstalling.

Library skills are baked into the EasyCC binary; new versions ship with EasyCC updates. Your edited copies aren't overwritten.

For how to author and share skills, see Skills how-to.

Your context and advanced

Two stacked sections.

Your context

A global text area of things Claude should always know about you — role, industry, style preferences, accessibility needs, tools you use or avoid. Sent with every chat in every agent.

How it composes with other context:

Source	Scope
Your context	Every chat in every agent (this panel)
`<agent>/.claude/CLAUDE.md`	Every chat in that agent
Learned memory	Background-updated, per-agent and shared

All three combine. Your context is explicit and immediate; CLAUDE.md is per-agent and you write it; learned memory is what Claude picked up.

Advanced

Control	Default	What it does
Connect to Claude	Anthropic (Claude.ai account)	Or API key, Custom gateway, Bedrock, Vertex, Foundry.
Token-efficiency features	All on	Expanded tool limits, skill model routing, read-once hints, MCP classifier, local proxy cache.
Open logs / agent folder / config / Reset window layout / Clear caches	—	Diagnostic actions.
Dev tools / Dev mode / Verbose logging	Off	For debugging issues with EasyCC support.
Telemetry	Off	Opt-in anonymous usage signal. See admin Security.
Experimental features	—	In-progress toggles. May break. Disappear as they stabilize.
Reset General / Reset layout / Reset all	—	Non-destructive — only resets settings, never files.

Common admin locks: provider pinned to gateway or Bedrock, local-proxy token-efficiency, telemetry.