Skip to main content
Workflows & Agents is where you invent, configure, and operate automations for the active project and end user. Confirm both switchers if something is missing — see External users switcher and Projects and API keys. Open Workflows.

​
Agent vs workflow

The product uses the same list for both. The help popover defines them as: Tip from the UI: if you are unsure, start with an Agent — you can add structure later. Conceptual detail: How Flowra works. Developer model: Workflows and agents.

​
List page

​
Header actions

  • Create Agent — blank agent for the end user selected in the top bar
  • Create by AI — opens Chat with a helper; you still review and save in the editor
  • Executions — global run history
  • Widgets — embed chat widgets for agents
There is no separate blank Create Workflow button on the list. Static workflows usually come from Create by AI, Templates clone, or other guided paths.

​
Filters

  • Search
  • Status: All / Active / Inactive
  • Type tabs: All | Workflow (fixed step-by-step) | Agent (conversational)

​
Cards

Each card opens the editor and typically shows:
  • Type badge (Agent or Workflow) and version when present
  • Active switch (activate / deactivate)
  • Description (or empty placeholder)
  • Trigger row — agents always show Chat, plus schedule or channel badges when configured; static with no triggers shows Manual
  • Last run, Runs count, toolkit avatars
Card menu (varies by type): If your plan is over the agents/workflows limit, create and edit may show Plan limit reached / Editing locked with a path to change plan.

​
Create an agent

Path: Create Agent (or Create Agent on the list). The form uses the same settings tabs as the editor (see below). Workspace is set from the current project; the agent is created for the selected end user.
  • Left: settings tabs
  • Right: live Flow visualization preview
  • Submit: Create Agent
Unsaved changes prompt when you leave. Prefer a small tool set from the start. You can add a website widget under Entry points so it is created when you save.

​
Edit an agent

Open an agent from the list. Header shows type, version, name, description, optional Sub-agents chips, and an Active switch when you can edit. Primary actions often include Edit with AI, Save Agent, and Chat / Chat with Agent. Menu may offer Add to templates / Remove from templates.

​
Performance

Performance summary: runs, success rate, last run, link to Full statistics. If there are errors, the UI may offer Fix errors with AI.

​
Agent settings (tabs)

Settings use five tabs. Counts on the tab labels reflect how many options you have configured.

​
Identity

  • Basic Info — Name, optional Description, Starter prompts (quick replies in chat / widgets; max 8), Avatar, Public (when your role allows — let other projects use this agent in their chat)
  • Start from a template — optional persona + Behavior pack (support, outreach, internal Q&A, and similar). Applying replaces persona, guardrails, escalation, canned replies, and usage caps; you can edit afterward
  • Persona — display name, role, tone, verbosity, language, emoji style, style notes
  • End-user identity — how visitors prove who they are on the widget, channels, and SDK (separate from toolkit OAuth like Gmail). Modes: Anonymous, Collect, or Required. Methods can include host HMAC/JWT, Email OTP, messaging-channel identity, SDK x-username, or a toolkit connection. HMAC and JWT keys live on the project End users tab — see External users switcher and Widgets and embeds

​
Intelligence

  • Agent Intelligence — System instructions, model picker, Temperature, Max tokens, Thinking when the model supports it, plus Advanced sampling
  • Output schema — optional structured fields for the agent’s final answer (API / integrations)
  • Token and cost estimate — rough scenarios (first turn, typical chat, long thread, custom); estimates only

​
Actions

Tools — pick what the agent is allowed to do. Open Select Tools (toolkits and MCP). Prefer a small set — the product may warn when too many tools are pinned. Full catalog: Toolkits and tools. Select Tools modal (summary):
  • Toolkit / built-in tools and External MCP
  • Normal pin vs Meta / discovery modes when offered
  • External MCP URL when needed
  • Apply selections back to the agent
Let visitors use my accounts (on the Tools card, after the agent is saved):
  • Off (default): each visitor connects their own account; the messaging channel is only for replies
  • On: selected apps run with your connected account for anyone who messages this agent on a channel or website — only enable apps you need
Skills — attach playbooks with Select Skills. Skills guide behavior; they do not automatically add tools. Full guide: Skills. Data access — tick Knowledge collections and Tables (database collections) this agent may use. Search tools for those collections are added automatically when the agent runs. For tables exposed to end users, the UI offers access-style presets (Private per user, Shared read-only, Submit only) and row/column controls. Create collections first under Database and Knowledge. Capabilities — common toggles (names follow the UI):
  • Computer Agent (may warn about credits)
  • Todo list
  • Helper sub-agent (built-in general-purpose)
  • Long-term memory
  • Conversation summarization
  • Retry on tool errors
  • Human approval for sensitive actions
  • Advanced: Tool selector, Model fallback, Model call limit / Tool call limit
Sub-agents — attach up to the max count of other agents from this project (coordinator / webhook behavior as the UI describes). The built-in helper is toggled under Capabilities and also appears on the flow.

​
Behavior

  • Guardrails — banned topics, never-promise items, refusal message, grounding mode, PII rules
  • Escalation — when to hand off to a human; on messaging channels, reply-only and notify-tool can pause auto-reply until an operator takes over from External users. Human approval here is tool HITL and does not pause the conversation
  • Canned responses — greeting, fallback, out-of-hours (guidance in the prompt, not a hard server lock)
  • Usage caps — messages per user per day, max turns / credits per conversation; enforced mainly on public inbound (channels, embed, SDK end users)

​
Entry points

Channels and triggers that start this agent. Full guide: Entry points and channels. Built-in (always on):
  • Dashboard chat — chat inside Flowra
  • Use as MCP — connect this agent to Cursor, Claude, and other MCP clients (Install on the card, or Install as MCP from the list)
Channels — website Chat widget, plus messengers such as Telegram, WhatsApp, Slack, Discord, Gmail, Instagram Business, Instagram Private, LINE, and Bale. People message there; the agent can reply. On create, you can add a widget that is created when you save. Full list: Entry points and channels. Automation — Schedule (cron + timezone; no messaging auto-reply) or Events (toolkit webhook/poll; no automatic reply — use tools for outbound actions). Use Add entry point (or Add channel / Add schedule or event). Security copy in UI: replies use your connected account; anyone who can message that channel can wake the agent.

​
Your accounts for testing

Card Your accounts for testing (agents) or Your accounts for this workflow (static):
  • Shows Testing as / Running as the current end user or project default
  • Connect or change toolkit accounts
  • Copy: connections are only for you; other people must connect their own accounts
See Connections. This is separate from Let visitors use my accounts under Tools.

​
Flow visualization (right)

Flow visualization shows nodes and edges for the current configuration (chat, channels, tools, skills, knowledge, computer agent, and related nodes). On agents this is largely a live preview of what you configured.

​
Edit a static workflow

Open a Workflow card. Shared pieces: Performance, accounts, templates actions, statistics, executions.

​
What differs from agents

Save with Save inputs when you are editing input defaults. Use Run to execute; use Production vs Test modes where the product offers them (test can mock tools depending on configuration).

​
Test and run

If a run pauses for approval, open the execution and use Approve / Reject (or resume flows the UI shows). Developer detail: Human in the loop.

​
Executions

​
Global list

Workflow Executions Filter by status (Running, Paused, Completed, Failed, Timeout), flow type (Agent / Static), and mode (Production / Test). Columns typically include execution id, workflow, status, type, mode, triggered by (Me / External user), progress, started, duration.

​
Execution detail

Open a row for:
  • Run information (started, completed, duration, mode, triggered by)
  • Progress (completed / failed / total nodes)
  • Errors per node
  • Input / output data
  • Resume when paused
Also use Tool execution logs for tool-level I/O.

​
Statistics

From a card or editor: Full statistics → /dashboard/workflows/[id]/statistics. Typical sections: run health, credits, tools, models, triggers, logs — with date range and Test/Production filters. Use this for debugging cost and failure patterns, not only single-run detail.

​
Widgets (agents)

From the list header Widgets, Create widget on an agent card, or Entry points → add a chat widget.
  • Widgets are for agent workflows in the project
  • After create, customize design and chat options (bubble/fullscreen, colors, uploads, quick replies, thread persistence)
  • Starter prompts come from the agent’s Basic Info
Full product guide: Widgets and embeds.

​
Templates

From the editor you can Add to templates or Remove from templates. When viewing a Templates listing, Clone for me (or equivalent) copies into your project under the current end user — then reconnect accounts and review tools.