Floo AI - Intelligent Assistant

Floo AI is the artificial intelligence assistant integrated into the platform. It uses language models (GPT-4, GPT-4o) to help with the creation, optimization, and debugging of workflows.

Capabilities

a) Workflow generation from description

Describe in natural language what you need and Floo AI generates the complete workflow with nodes and connections.

Example:

"When I receive an order via webhook, check the stock by calling my API,
if there is stock send a confirmation email, if not notify via Slack"

Floo AI automatically generates:

• Webhook node as trigger
• HTTP node to query the stock API
• Decision node to evaluate the response
• SendMail node on the true branch
• Slack node on the false branch
• Connections between all nodes

b) Node suggestions

When describing a task, Floo AI suggests the most appropriate nodes. For example, if you type “send data to a PostgreSQL database”, it suggests the sqlQueryPostgres node with the relevant configuration.

If the AI service is unavailable, the system uses local keyword search as a fallback.

c) Optimization of existing workflows

Floo AI analyzes an existing workflow and suggests improvements:

• Redundant nodes that can be removed
• Parallelization opportunities
• Improvements to decision structures
• More efficient use of iterators

d) Error analysis

When a node fails, Floo AI analyzes the error in context and suggests:

• Probable cause of the error
• Recommended solution
• Correct node configuration
• Also offers quick suggestions without consuming AI tokens

e) Conversational chat

An integrated chat where you can ask questions about:

• How to configure a specific node
• Which module to use for a task
• How to solve a problem in your workflow
• Search for modules by functionality

f) Workflow editing from the chat (Edit Mode)

Edit Mode allows you to modify a workflow open in the editor directly from the chat panel, describing changes in natural language. Floo AI applies the changes to the existing graph without replacing it from scratch.

Note

On-premise installations only

Environment variables are set on the server where Floogos runs, so they apply to on-premise installations. On Floogos Cloud you don’t have access to them — these values are managed by the Floogos team. If you need one changed for your account, get in touch with support.

This feature is experimental and must be explicitly enabled with the environment variable VITE_AI_WORKFLOW_EDIT=true. If the variable is absent or set to false, the chat panel works in read-only mode as usual.

How to enable it

Add the following variable to the frontend environment file:

VITE_AI_WORKFLOW_EDIT=true

How to use it

1. Open a workflow in the editor.
2. Open the Floo AI chat panel.
3. Activate the Edit Mode toggle (pencil icon in the panel header).
4. Optionally, select a node in the editor to give the assistant context.
5. Describe the change you want to apply in natural language.
6. The assistant proposes the change with a summary of what it will modify.
7. Click Apply to editor to see the changes reflected in the graph.
8. When satisfied, save the workflow with Ctrl+S as usual.

Examples of valid instructions

"Add a sendMail node after the decision when the condition is true"
"Connect the router to the webresponse node with a 404 when the method is not POST"
"Change the subject field of sendMail_1 to 'Order approved'"
"Remove http_2 and reconnect the decision directly to end"

Editor behavior when applying changes

• New nodes appear with an animated green border for 3 seconds.
• Modified nodes appear with a blue border for 3 seconds.
• The previous state is saved in a local history (up to 10 states). You can undo the last AI change by clicking Undo last change in the chat panel.
• Changes are not persisted until the user manually saves the workflow.

Selected node as context

If you select a node in the editor before writing the instruction, the assistant uses it as a reference. This allows instructions like “change the HTTP method of this node to POST” without specifying the node ID.

Token consumption

Each edit consumes tokens equivalent to a workflow generation, since the model processes the full graph plus the instruction. Check your plan limits at AI Limits.

AI API Endpoints

Endpoint	Method	Description
/api/ai/generate-workflow	POST	Generate workflow from description
/api/ai/edit-workflow	POST	Edit existing workflow from natural language instruction
/api/ai/suggest-nodes	POST	Suggest nodes for a task
/api/ai/analyze-error	POST	Analyze a node error
/api/ai/chat	POST	Chat with the assistant
/api/ai/optimize-workflow	POST	Optimize existing workflow
/api/ai/modules	GET	List all available modules
/api/ai/modules/search/:query	GET	Search modules
/api/ai/categories	GET	List module categories
/api/ai/token-estimate	POST	Estimate tokens before executing
/api/ai/usage/stats	GET	Monthly usage statistics
/api/ai/usage/limits	GET	Plan limits vs current usage

Token Consumption

• Each AI operation consumes tokens from the configured model
• The system estimates tokens before executing to avoid overages
• Usage is tracked per client and per operation type
• Limits depend on the contracted plan (queryable at /api/ai/usage/limits)