Workflow Assistant

The Workflow Assistant is an AI-powered chat panel that lives inside your workflow editor. Instead of dragging and configuring nodes manually, you can describe what you want in plain English and the assistant builds it for you.

Open it by clicking the floating button on the right edge of the editor.

What It Can Do

The assistant has direct access to your workflow canvas. It can:

• Add nodes — create any node type and place it on the canvas
• Connect nodes — wire outputs to inputs with proper handle mapping
• Configure nodes — set properties like asset symbols, LLM prompts, conditions, and more
• Delete and reorganize — remove nodes, rename them, reposition them on the canvas
• Execute workflows — trigger a run and monitor the results
• Debug failures — pull execution logs, analyze errors, and suggest fixes

Every action the assistant takes appears as a live update on your canvas — you see nodes being added and connected in real time.

How the Agent Works

At a high level, each message goes through a simple loop:

1. Gather current context (your workflow graph, variables, and recent execution data)
2. Decide which built-in tools to call (if needed)
3. Apply tool results to the workflow or fetch supporting data
4. Return a final response with actions and guidance

This means the assistant is not just "chatting" — it is planning and taking structured actions against your workflow.

Tool Categories

The assistant uses a fixed internal toolset. It does not have unrestricted access to your account or environment.

Conversation Compaction

Long chats are automatically compacted to keep responses fast and relevant.

• Early messages are summarized into a dense context note
• Recent turns are kept in full detail
• Current request and latest workflow state are always prioritized

You may occasionally see a status like "Summarizing earlier conversation..." when this happens.

Safety and Data Boundaries

To protect users and credentials, the assistant follows strict boundaries:

• It only uses scoped workflow tools provided by NickAI
• It cannot reveal hidden system prompts or private backend configuration
• Credential values (API keys, secrets) are not surfaced in assistant responses
• Some actions still require manual user updates (for example, fixing invalid/expired credentials)

These guardrails are designed so you can understand what the assistant is doing without exposing sensitive internals.

Building with Natural Language

Just describe what you want. The assistant understands trading concepts and workflow logic. Watch it build a sentiment trading bot from a single prompt:

The assistant creates all nodes first, then wires them together, then configures properties — just like you would manually, but in seconds.

Actions

When the assistant modifies your workflow, each change appears as an action pill in the chat showing exactly what happened:

Each action shows a status indicator — spinning while executing, checkmark on success.

Debugging

When a workflow fails, the assistant can diagnose the issue:

1. It retrieves the execution logs and per-node inputs/outputs
2. Analyzes where the failure occurred
3. Explains the root cause in plain language
4. Suggests or applies a fix directly

Auto-Iterate

Enable Auto-Iterate to let the assistant automatically fix and retry failed workflows:

1. You click Run (or the assistant triggers execution)
2. If the workflow fails, the assistant analyzes the error
3. It modifies the workflow to fix the issue
4. Reruns automatically
5. Repeats up to 5 times or until the workflow succeeds

Enable auto-iterate from the toggle in the chat input area. Your preference is saved across sessions.

Voice Input

Click the microphone icon to speak instead of type. The assistant uses your browser's speech recognition to convert voice to text.

• Automatic silence detection — stops recording after 2 seconds of quiet
• Real-time visualization — audio level indicator shows recording is active
• Works in Chrome, Edge, and Safari

Attachments

You can attach files to your messages:

• Images — share screenshots of charts, error messages, or UI issues (up to 20MB)
• Documents — attach PDFs, text files, or spreadsheets for the assistant to reference (up to 20MB)

Useful for sharing chart screenshots for analysis, or pasting error logs from external systems.

Models

The assistant supports multiple LLM providers through a unified interface. Switch models from the dropdown in the chat input area.

Some models support extended thinking — you'll see a collapsible "Thought process" block showing the assistant's reasoning before its response.

How It Understands Your Workflow

The assistant receives full context about your current workflow:

• All nodes — types, positions, labels, configurations
• All connections — which outputs connect to which inputs
• Node schemas — what each node type accepts and produces
• Execution history — recent run results, errors, and per-node outputs
• Variables — workflow and global variables you've defined

This context means the assistant can reference specific nodes by name, understand data flow between them, and make targeted fixes.

Next Steps

• Workflows — learn the visual editor fundamentals
• Executions — understand how workflows run and how to debug them
• Nodes — explore all available node types the assistant can use
• Credentials — set up the API keys your workflows need