Settings

Settings open as a single screen; which page you see is chosen in the sidebar. There are ten sections, always in the same order: Account, Workspace, Models, Secrets, Integrations, Assistant MCP, Widget, White-label, Usage, Plan & billing.

Account

Your profile (name, email, sign-up date), the interface language toggle and a card for the current workspace with its name and slug. The “Sign out” button is here too.

Workspace

The workspace name and who belongs to it. The name is edited by clicking it and saves immediately; the slug is set at creation and never changes — it lives in addresses and links.

Below it is the member table with the columns “Member”, “Role” and “Actions”, and next to it an “Add a member” form: email and role. There are three roles, and they decide what a person may do inside this workspace:

The workspace owner is a row of their own: they cannot be removed and their role does not change. They also do not occupy a seat in the plan’s member quota, so a plan “for N members” means N invited people plus the owner.

Archiving a workspace and its credit budget are account owner operations rather than workspace-admin ones, and they are not in this section. Archiving undeploys every workflow of the workspace (the Telegram webhook is removed, the schedule stops) but keeps the public tokens, so restoring brings back the same addresses — though it does not re-deploy anything by itself.

Models

Every model available to LLM node dropdowns, grouped by source:

• registry — the platform catalog;
• workspace — models from your own connections (your provider key);
• chain — fallback chains, several models with failover between them.

The switch on the right controls whether the model shows up in the editor. When a model has a tariff, its price in credits per 1000 tokens is shown next to it — input and output separately. Chains have no switch: their members and order are edited under Integrations.

Secrets

Name/value pairs belonging to the workspace. Templates refer to a secret as {{ secret.NAME }}, so name them the way environment variables are named: OPENAI_API_KEY.

A value is never shown again once saved. When editing, an empty value field means “keep the current one”. Deleting a secret does not break workflows on the spot: references to it start resolving to an empty string, which usually surfaces as a rejection from the external service. Creating, changing and deleting secrets requires the workspace admin role.

Integrations

Connections to external services and model providers: Email (IMAP/POP3 + SMTP), WebDAV, CalDAV, MCP, an OpenAI-compatible LLM, a Telegram bot, a model fallback chain, Bitrix24, amoCRM, VK and VK Teams. The kind is chosen at creation and never changes afterwards — the form’s fields depend on it.

Email, WebDAV and CalDAV get a “Check connection” button; it probes the saved connection, not the unsaved edits in the form. The other kinds have nothing to probe — they fail loudly on first use anyway.

Deleting a connection is confirmed: workflows that reference it stop working afterwards. Changes require the workspace admin role.

Assistant MCP

The key an external AI agent (Claude Code, for instance) uses to drive the platform over MCP. The page carries the endpoint URL, a ready-made CLI command and an .mcp.json block, each with a copy button.

There is one key per workspace. A freshly minted token is shown once, right after it is generated; afterwards you only see its first and last characters, the last-used date and the status. “Rotate key” issues a new one in place of the old, “Disable” removes access entirely.

Below that sit twenty permissions grouped by subject (docs, connections, validation, workflows, executions, knowledge bases, files, database, chat, persistent state). Only the reads are on by default: the agent cannot save, run, deploy or delete anything until you turn on an explicit switch. Toggles apply immediately. Managing the key and its permissions requires the workspace admin role.

Widget

The secret used to sign end-user JWTs, plus the snippet that embeds the chat widget on your site. That secret is also revealed only once, has a rotate button and a last-used marker. Embedding details live on the chat widget page.

White-label

Your own logo, brand name and accent color for the public chat and the widget, plus a “Hide Powered by Flow” switch. The section works when the workspace owner’s plan grants white-label; otherwise the inputs are disabled and an explanatory note is shown.

Usage

The consumption dashboard: plan and subscription status, a bar of credits spent, quotas for the knowledge base, files, database and vector points, a breakdown of spend by event type (LLM tokens, media generation, web search, workflow runs and so on), a bar chart of the last 30 days, and a list of knowledge bases with a “Stale” badge for any that have not been updated in over 30 days.

Plan & billing

Your current plan, subscription status and credit balance, and below them the available plans and credit packs with prices and a “Buy” button. A purchase either sends you to the payment provider’s page or is applied in place when the provider allows it; the balance and the “Current” marker refresh by themselves afterwards. If the operator has disabled self-service billing, the buttons are inactive and a note says so.

Where to next