Skip to content

FlowSettings

allowAgentToolLoop: boolean

Org-level master switch for the in-node AI tool loop (an ai node calling bound adapters mid-inference). When off, every tool-loop node fails at run time regardless of its own allowToolLoop. Defaults true.


allowAutonomousRun: boolean

Org-level master switch for autonomous runs (auto-accept + re-run the monitor’s fixes without a human gate). When off, an autonomous request is refused / downgraded to manual review. Defaults true.


allowCloudAi: boolean


allowLocalAi: boolean

Gate for the local OpenAI-compatible provider (on-device server). Independent of allowCloudAi because a localhost call isn’t egress.


confirmDestructive: boolean

Per-step destructive-action confirmation gate. When on, a run pauses before any node that performs a destructive operation and asks the user to confirm or cancel. Defaults to true.


optional hubUrl?: string

Base URL of the Flow Hub service (registries + template publishing). Empty/absent = offline mode (embedded catalogs only).


optional llamaParams?: Record<string, LlamaParams>

Per-model llama-server load settings, keyed by catalog model id.


optional llamaServerBinary?: string

Path to the llama-server binary flow-studio launches to host an LLM.


optional llamaServerModel?: string

Path to the LLM the managed server loads.


optional localAiBaseUrl?: string

Base address of the local server, stored as an origin (e.g. http://127.0.0.1:1234). The runtime appends /v1/chat/completions and /v1/models. A full URL is also accepted and normalised.


optional localAiModels?: string[]

Model ids discovered from the local server’s /v1/models, cached so the node inspector dropdown can offer them. Populated by Test connection.


maxAgenticIterations: number

Safety cap on auto-accepted fixes in one autonomous run; one budget shared by the backend convergence loop and the frontend monitor auto-fix loop.


maxAgenticSeconds: number

Wall-clock ceiling (seconds) for one autonomous convergence run; 0 = unlimited (autonomous run budgets).


maxAgenticTokens: number

Cumulative token ceiling (the executed flows’ AI nodes) for one autonomous convergence run; 0 = unlimited.


piiRules: CustomPiiRule[]

Admin-authored PII redaction rules layered on the built-in sanitizer. Applied to every AI-node input before inference.


providersEnabled: Record<string, boolean>


sandboxLocalLlm: boolean

OS-level isolation for the managed reasoning server. When on (macOS), llama-server runs under a sandbox-exec profile so loopback-only is enforced by the OS. Off by default.


schedulerEnabled: boolean

Master switch for background scheduled runs.


schedulerPollSecs: number

Poll interval in seconds for background scheduler loops.


theme: UiTheme