Settings

Settings is where you tell C5 how to behave. This page is a map, so you know which area you want.

You do not need to change most of these. C5 works sensibly out of the box.

The areas #

The ones worth setting early #

Integrations #

You have to set this up before anything works. Add at least one model and its key. See Model providers.

Budget #

Set a spending limit on your first day. It is the cheapest mistake-insurance available. The Budget tab is also where you track progress toward self-improvement milestones like the Retrospective Loop (50 tasks) and Reward Model Promotion (100 tasks). See Budgets & costs.

Autonomy #

Decide how much your agents may do on their own. Start cautious. As you learn what C5 does well, loosen it.

See Permissions.

Guardrails #

Set boundaries on what your agents may do, where they can connect on the web, and what data gets redacted before leaving your computer. See Guardrails.

Locations #

Tell C5 which folders to bring documents in from, and where to put finished work. Point it at the folders you actually work in, so imports and deliverables land somewhere you expect.

How long C5 keeps things #

Storage is where you decide how much history to hold. Two of the sliders there are worth understanding, because they control records you may need after the fact.

Audit Trail Retention covers the security record of who did what: sign-ins and account changes, admin actions, attempts C5 blocked, and changes to encryption. C5 removes it a whole day at a time, and only once the last entry from that day has aged out, so nothing inside your window is ever cut short. See Audit trail.

This slider also covers the tool decisions in the same view — each time an agent was allowed or refused a tool. Those are tidied on the same nightly pass, but with one difference worth knowing:

One thing in the Audit view is still not covered: the system notifications merged into the same list have no retention setting, and nothing ages them out. The Source column is what tells the three kinds of row apart.

System Events Retention covers C5's own operational record: coordinator failovers, gateway timeouts, shutdowns, steps that had to be abandoned and requeued, and the adjustments C5 makes to its own thresholds and prompts as it learns. This is what the Monitor screens read when they tell you how the platform itself has been behaving.

Both are cleared out once a day, late in the evening, a few minutes ahead of the nightly database backup — so your backups hold the tidied-up version rather than a copy of what you asked to delete. They tidy differently: the audit records go a whole day at a time, for the reason described on the Audit trail page, while tool decisions and system events are removed one at a time as each passes its cutoff.

If C5 is asleep or switched off at that moment, nothing is lost. The next night removes everything that was due tonight as well.

If you need records for longer than you want to store them, export them instead of raising the slider — see Audit trail.

Alerts #

Alerts tell you when something needs you. Worth turning on:

• A task is blocked and waiting for your answer
• A schedule failed
• You are close to a spending limit
• Something was blocked for security reasons

Without alerts you will find out when you next open C5, which might be tomorrow.

Access & user permissions (RBAC) #

Settings → Access (labeled Security & Access Configuration) is where administrators configure what standard (non-administrator) teammates are permitted to see and do throughout C5.

Administrator accounts always retain full view, edit, and execution privileges across all pages, features, and system controls. For standard teammates, administrators can fine-tune access using two levels of control on every feature:

• User View: Controls whether standard users can see a feature, page, or action button. When turned off, the navigation tab, sidebar link, or button is hidden from their interface.
• User Edit: Controls whether standard users can modify records, trigger active pipelines, or perform actions. When edit is turned off, the interface switches to read-only (disabling save buttons, edit forms, creation modals, and action menus).

At the top of the page, a global Killswitch banner allows an administrator to immediately halt all autonomous agent execution across the entire platform. When active, background processes pause and the controls dim.

The RBAC grid #

The permissions grid organizes controls into three columns:

1. Feature locks (left column)

Granular controls over specific high-impact capabilities:

User self-help: Missing Restart, Quit, or Start buttons If you are a standard user and do not see the Restart C5, Quit C5, or Start C5 buttons in your sidebar menu (⋯), or the Start C5 button on connection error screens, your administrator has kept Sidebar: Restart, Quit, Start C5 in its default Off position.

When an administrator enables this toggle, the permission automatically synchronizes to your browser (localStorage) the next time you sign in, unlocking the server lifecycle controls.

2. Page locks (middle column)

Visibility and edit capabilities for primary application sections:

3. Page locks & hierarchy (right column)

Advanced workspaces, agent fleets, and system configuration:

Hierarchical cascading #

The master locks for Library, Ops, and Labs simplify management:

• Turning a master View toggle off immediately hides all nested sub-tabs for standard users.
• Turning a master Edit toggle off locks all nested sub-tabs into read-only mode.
• You can still selectively enable or disable individual child tabs while keeping the parent active.

OpenClaw gateway restrictions #

Below the RBAC grid, the Access tab provides direct configuration for the OpenClaw gateway security boundary:

• Agent Capabilities: Enable or disable agent-to-agent delegation, shell execution, browser automation, sub-agent spawning, and MCP tool access.
• Filesystem Boundaries: Confine agent read/write operations strictly to designated workspace paths.
• Network Allowlist: Enforce explicit host allowlists for outbound network traffic.

Saving and synchronization #

When an administrator clicks Save on the Access tab:

1. Permissions update immediately in the central configuration database.
2. Active user sessions update in real time.
3. For offline-capable controls (like the Sidebar: Restart, Quit, Start C5 toggle), the client browser automatically synchronizes the permission into local browser storage (localStorage) upon sign-in. This ensures offline fallback screens know whether to present server restart buttons even when the backend is unreachable.

Starting, restarting and quitting C5 #

Settings → System carries the controls that act on the server, and the same options sit at the bottom of the sidebar Options menu (the ⋯ button) so you can reach them without leaving what you are doing. Both Restart and Quit ask you to confirm first — they act on the server everyone on this machine is using.

Note on permissions For standard (non-admin) team members, the visibility of Restart, Quit, and Start buttons is controlled by the Sidebar: Restart, Quit, Start C5 toggle under Settings → Access. By default, this is turned off to prevent accidental server shutdowns.

Two settings sit beside them and are often confused:

• Start when I sign in registers C5 with your operating system, so it is already running the next time you log in.
• Restart automatically if it stops unexpectedly covers crashes only. An intentional quit — this button, Ctrl+C, or growther stop — is respected and stays stopped. If it resurrected an explicit quit, you could never stop C5 at all.

When C5 is not running #

When the server is stopped or offline, the interface adjusts automatically:

• In Settings → System, Restart is disabled, and the Quit panel transforms into a green Start C5 panel.
• In the sidebar Options menu (⋯), Restart and Quit are replaced by a centered green Start C5 button.
• On the offline fallback page and connection error modal, an emerald Start C5 button appears below the details view.

Clicking Start C5 invokes your computer's built-in growther://start protocol link to launch C5 (or wake its background service). While it starts, the button shows a spinning indicator with Starting…. Once C5 is running, your tab reconnects on its own within a few seconds and the online controls return.

You can also start C5 at any time from the "Growther.ai C5" desktop shortcut, or by running growther start in a terminal.

Saving changes #

Changes save when you click Save and apply to new work right away. Work already running finishes under the settings it started with, so nothing shifts mid-task.

If you break something #

Settings are just values you can change back. Note what you changed, set it to what it was, and save.

If you are not sure what you changed, the Audit view in Ops covers some of it: changes to what tools an agent may use, and to search providers and integrations, are recorded there with a timestamp. An ordinary Save on the other settings areas is not recorded there — so when a setting matters, note what it was before you change it.

If C5 will not start at all after a change, see Common issues.

Where to go next #

• Config files — how C5 manages c5.yaml on disk.
• Model providers — connect cloud and local models.
• Guardrails — safety limits and network allowlists.