NxAgent User Guide
NxAgent is an AI-powered assistant embedded directly into the Koodisi workspace. You remain responsible for reviewing generated content, testing changes, and deciding what enters an application.
Overview
You can ask NxAgent questions, instruct it to build workflows, generate schemas, or investigate integration issues — all without leaving your canvas.
With NxAgent, you can:
- Ask questions about your applications, workflows, and schemas in plain language.
- Design new workflows and schemas by describing what you need in natural language.
- Triage integration health by querying live metrics and performance patterns.
- Attach context from your existing resources to make responses precise and relevant.
Key Concepts
| Term | What it means |
|---|---|
| Chat Mode | Controls NxAgent's focus: Ask (Q&A), Design (generation), or Triage (monitoring). |
| Response Style | Controls tone and depth: Concise, Learning, Explanatory, or Formal. |
| Context | Existing workflows or schemas you attach to inform NxAgent's response. |
| Skill | A knowledge file (.md or .zip) uploaded by your org admin to extend NxAgent's domain awareness. |
| Thread | A single conversation session. Each new thread starts fresh with no prior history. |
Benefits
Here is how NxAgent improves your day-to-day work:
- Speed up design: Instead of manually building workflow logic node by node, describe what you want and NxAgent drafts it instantly.
- Get instant answers: Instead of searching docs or asking colleagues, ask NxAgent directly about your specific workflows and configurations.
- Diagnose faster: Instead of manually scanning logs and metrics, Triage mode surfaces the most important issues in plain language.
- Personalized depth: Instead of one-size-fits-all answers, select a response style that matches your level of expertise or the situation.
Write an effective request
State the outcome, relevant constraints, and how you will judge the result. Attach only the application items needed for the task.
Create a webhook workflow that validates an employee record, rejects requests without an employee ID, and returns a clear error. Use the attached schema. Do not change existing workflows. Show me the proposed changes before applying them.
For investigation, include the environment, time window, observed symptom, and expected behavior. Do not include passwords, tokens, or customer data in a prompt.
Keep people in control
- Review Preview and Diff before accepting a generated change.
- Test generated workflows and schemas with representative, non-sensitive data.
- Keep Auto accept off for shared applications, unfamiliar tasks, and release-critical work.
- Treat generated explanations as guidance. Verify operational and security decisions against the application and your organization's policies.
- Use the smallest useful context. Remove unrelated workflows and schemas before sending the request.
Step-by-Step Guide
Task 1: Opening NxAgent
Goal: Access the NxAgent chat panel from anywhere in the workspace.
Steps:
- From any application workspace, locate the Chat With AI panel on the right-hand sidebar.
- Alternatively, if you see the "Chat. Design. Build." card on a landing screen, click Try NxAgent →.
- The Chat With AI panel opens, showing the empty-state screen with suggested example prompts.
Result: NxAgent is open and ready for your first message.
Task 2: Choosing a Chat Mode (Ask, Design, Triage)
Goal: Select the right mode for your task before starting a conversation.
NxAgent has three modes. Select from the mode dropdown in the bottom-left of the input bar.
| Mode | Best for | Example prompts |
|---|---|---|
| Ask | Questions about existing workflows, schemas, errors, and configurations. | "Explain this workflow." / "Why is this mapper step failing?" |
| Design | Generating brand-new workflows and schemas from a description. | "Create a workflow that processes webhooks and stores data in a database." |
| Triage | Investigating operational health, metrics, and recurring integration problems. | "Are there any critical issues right now?" |
Steps:
- Click the mode dropdown in the bottom-left of the input bar (shows current mode, e.g. Ask).
- Select Ask, Design, or Triage.
Important: The mode is locked once you send the first message in a thread. To switch modes, click + (New Thread) in the panel header.
Result: NxAgent is configured to respond according to the selected mode.
Task 3: Setting a Response Style
Goal: Control how NxAgent explains its answers.
Steps:
- Click the pen icon (✏) in the input toolbar.
- Choose from the four available styles:
| Style | Best for |
|---|---|
| Concise (Default) | Fast, direct answers. Best for experienced users. |
| Learning | Includes reasoning and context. Best for understanding new concepts. |
| Explanatory | Detailed, structured breakdowns. Best for deep dives or sharing outputs. |
| Formal | Professional tone. Best for stakeholder-facing outputs. |
- The checkmark indicates the active selection.
Result: All responses in the current thread will use the selected style.
Task 4: Adding Context to a Prompt
Goal: Attach an existing workflow or schema so NxAgent answers specifically about your data.
What you need: An existing workflow or schema in your application.
Steps:
- In the input bar, click the attachment icon (📎), or drag and drop a workflow or schema from the left-hand sidebar directly into the chat input area.
- A context tag showing the resource name appears above the text input.
- To remove a context item, click the ✕ on its tag.
- Type your prompt and send.
Note: Context can only be added before the first message in a thread. Once a conversation starts, the context is locked.
Result: NxAgent references your attached resource to give a specific, targeted answer.
Task 5: Sending a Message and Reading Responses
Goal: Send a prompt and understand the types of responses NxAgent can return.
Steps:
- Type your query in the "Ask NxAgent..." input field, or click an example prompt chip to pre-fill it (e.g., "Explain this workflow", "Add error handling").
- Press Enter or click the send arrow (→) button.
- NxAgent processes your message. Responses take several forms:
| Response Type | What you see | Available action |
|---|---|---|
| Text / Markdown | Structured text with formatting, lists, or tables. | Read and copy. |
| Workflow | A visual preview of a generated workflow node graph. | Click Accept to add it to your canvas. |
| Schema | A structured JSON schema preview. | Click Accept to import it into your Schema Registry. |
| Chart (Triage) | An inline bar, line, area, or pie chart alongside written analysis. | Review; maximize the panel for a larger view. |
- After a response, you can:
- Type a follow-up message to continue the conversation.
- Edit a past message by hovering over it and clicking the edit icon — the conversation resets to that point.
- Provide feedback with thumbs up/down below any response.
- Re-run a query using the refresh option in the feedback panel.
Result: You receive a targeted answer, generated artifact, or operational insight from NxAgent.
Task 6: Reviewing and Applying Generated Changes
Goal: Verify proposed application changes before NxAgent writes them to your application.
When NxAgent proposes changes, the preview can include workflows, resources, schemas, global variables, and application-level operations.
Steps:
- Open the generated-change preview.
- Select an item from the file list.
- Use Preview to inspect the complete proposed artifact or Diff to compare it with the current version.
- Select Accept to apply only the selected item, or Reject to leave it unchanged.
- Use Accept all only after reviewing every proposed item.
Important: Accepting a deletion or replacement changes the selected application. Review the application context and file path before confirming.
If Auto accept is enabled, NxAgent applies proposed changes without waiting for file-by-file approval. Keep it disabled for unfamiliar prompts, shared applications, and Production-related work.
Result: Only the changes you approve are applied to the application.
Task 7: Managing Conversation History
Goal: Browse past conversations or start a clean new thread.
To start a new thread:
- Click the + (New Thread) icon in the panel header.
- The panel resets. Prior conversations are saved and accessible from History.
To browse past conversations:
- Click the History icon (clock) in the panel header.
- A list of past chat sessions is shown.
- Click any session to reload it and continue from where you left off.
To maximize the panel:
- Click the expand icon in the header. Especially useful for reviewing Triage charts or long workflow previews.
Result: You can manage multiple conversations and revisit AI outputs at any time.
Troubleshooting & FAQs
Common Issues
Problem: NxAgent gives generic answers, not specific to my workflow.
- Solution: Attach your workflow or schema as context before sending. Drag it from the left panel directly into the chat input area.
Problem: The mode dropdown is greyed out.
- Solution: The mode locks once a conversation starts. Click + (New Thread) and choose your mode before sending the first message.
Problem: There is no "Accept" button after a response.
- Solution: The Accept button only appears for workflow or schema generation responses. Text answers do not have an Accept action.
Problem: Triage mode is not reflecting real-time data.
- Solution: Check the environment selector in the input bar. Ensure you have selected the correct environment (Development or Production).
Frequently Asked Questions
Q: Can NxAgent access a specific application's data? A: Yes. Use the application context selector (top-right of the panel) to switch between applications, and attach specific workflows or schemas as context before prompting.
Q: Are my conversations saved? A: Yes. All threads are saved automatically and accessible from the History panel (clock icon in the header).
Q: Can I edit a message I already sent? A: Yes. Hover over any past message and click the edit icon. The conversation rolls back to that point so you can revise and resend.
Q: How do I make NxAgent aware of my organization's internal systems? A: Your Organization Administrator can upload AI Skills under Settings → Organization → AI Skills. See Governing AI for details.
Maintained By: Koodisi Support Team