Permissions and local data

Raticode keeps its working state on your machine and makes tool access visible. The provider you choose still determines where its requests go and how its permissions work.

Separate the permission layers

Layer	What it controls
Project selection	The workspace used by the studio and project tools
Rem resources	Shell, web, skills, MCP, and selected application tools exposed to a turn
Provider permissions	The provider's approval mode and native execution policy
Workflow preflight	Readiness of declared executables, paths, credentials, and contracts
Organization or device grants	Access allowed to a specific employee or trusted remote request

Disabling Rem's shell resource does not disable a separately configured MCP server's abilities. Provider-native tools can have their own policy. A project picker is not an operating-system sandbox.

Read-only sessions can inspect supported state but cannot mutate through application management tools. An editable session still needs an instruction to perform the requested work; selecting a resource is not the same as asking it to execute a workflow.

Paths and side effects

Registered Rattish workflows resolve relative paths from their project root. Standalone files resolve relative paths from their source directory. Absolute paths, home expansion, and parent traversal are supported subject to the node and operating-system checks, so review paths before execution.

Commands and agents can change files, call services, and launch processes. Preflight does not simulate those effects, and Stop does not roll back completed changes. Review imported workflows and preserve recoverable work before running them.

Credentials

Authenticate coding-provider CLIs through their own supported login flow. API-backed providers need their configured credentials. Workflows can use named environment-backed references such as secret["SERVICE_TOKEN"]; do not paste the actual value into source, reports, or screenshots.

Device identities and organization secret values have dedicated platform credential storage. Organization tools can grant a secret reference without storing the value in task text. Ordinary workflow environment variables are not automatically an encrypted vault, and the whole project or chat archive is not encrypted at rest by this feature.

Stored data

Project workflow source and editor metadata live under the project's .raticode/ directory. App state uses Raticode's data directory by default. The folder keeps its original gofer name so existing data stays in place:

Platform	Default backend data location
Linux	$XDG_DATA_HOME/gofer, or ~/.local/share/gofer
macOS	~/Library/Application Support/gofer
Windows	%APPDATA%\gofer

Some CLI commands allow a custom data directory. The desktop's browser-profile storage, including working chat state, is separate from backend files. Swarms and organizations maintain their own persisted state.

An enabled conversation archive creates a separate copy in the chosen folder. Second Brain indexes a configured knowledge root. Back up each location you rely on; deleting one copy does not necessarily delete another.

Network behavior

Raticode does not require a Raticode cloud account. Provider requests, enabled web search, remote MCP servers, model downloads, Tailscale if you turn it on for the phone, and explicit HTTP workflow nodes can still use the network. Local state does not make every feature offline.

See Provider setup, Tools and resources, and Troubleshooting.