---
title: "Debugging"
url: "https://help.magicblocks.ai/en/article/debugging"
language: "en"
last_updated: "2026-03-16T09:52:15.339Z"
---

# Debugging

**Debugging** helps you identify, analyze, and fix issues in your AI’s logic, responses, or goal paths.  
It’s your behind-the-scenes view of what really happened inside a conversation — why your Agent responded a certain way, which block or rule it triggered, and how it processed user input.

Think of Debugging as the **AI’s black box recorder** — every interaction, every decision, and every trigger can be traced back to improve accuracy and reliability.

## **Why It Matters**

When your Agent doesn’t behave as expected — missing a goal, giving a weak response, or repeating lines — Debugging helps you uncover exactly *why*.

It allows you to:

Trace the **logic path** the AI followed.

Verify which **Brain source** or **Journey block** was used.

Identify **missing or incorrect context**.

Catch **broken or misfired conditions** before users do.

With Debugging, you’re not guessing what went wrong — you’re seeing it clearly.

## **Where to Find It**

You can open Debugging from:

**Sessions → Reviewing**

Click the **🤖 robot icon** next to any AI message.

The **Troubleshooting drawer** will open, showing technical details about how that specific message was generated.

## **What You’ll See in Debugging Mode**

| Section | Description |
| --- | --- |
| **Message Context** | The system’s internal understanding of user intent and conversation state before the AI responded. |
| **Knowledge Source** | Which Brain document, collection, or Q&A the AI used to form its answer. |
| **Journey Path** | Which Journey block or condition was triggered (e.g. “Hook → Align → Conversion”). |
| **Conditions Matched** | A list of IF/THEN logic checks and their outcomes (matched or skipped). |
| **Version** | The specific Agent version that handled the message — helpful when testing iterations. |
| **Token Usage** | (Advanced) How many tokens or steps the AI used to generate the response. |
| **Error Logs** | Warnings about missing context, failed goal triggers, or invalid actions. |

## **Common Debugging Scenarios**

| Problem | Possible Cause | Debugging Fix |
| --- | --- | --- |
| **AI gave irrelevant response** | Wrong or outdated Brain source | Reassign Knowledge collection or update context. |
| **Goal not triggered** | Condition mismatch or missing path | Check Journey logic; ensure block transition is defined. |
| **AI repeated messages** | Recursive prompt or memory loop | Reset memory or add stop conditions. |
| **No response generated** | Invalid Action or API error | Check integration (e.g., webhook or calendar task). |
| **AI tone inconsistent** | Conflicting Persona overrides | Verify Persona per block and global tone settings. |

## **How to Debug a Session**

**Open any active or completed session** in the Sessions tab.

Identify the message that needs inspection.

Click the **🤖 icon** beside it — this opens **Troubleshooting Mode**.

Review:

**Context** – what the AI knew before replying.

**Source** – where it pulled its answer from.

**Logic** – which Journey block or action triggered.

**Conditions** – which IF/THEN checks were matched.

Use this insight to adjust your **Brain**, **Journey**, or **Persona**.

Recreate the same conversation (using the **Recreate Session** button) to verify your fix.

## **Pro Debugging Techniques**

| Technique | Description | Benefit |
| --- | --- | --- |
| **Version Comparison** | Compare v1 vs v2 session paths to test AI improvements. | See measurable progress in tone, clarity, and conversions. |
| **Tag Failed Goals** | Use session tags like “Goal Missed” or “Debug” for tracking. | Organize your optimization workflow. |
| **Recreate Session** | Replay the same user messages after edits. | Confirm that your changes solved the issue. |
| **Cross-Link with Knowledge** | Check if the AI missed relevant content in Brain. | Fill Knowledge gaps to avoid future errors. |

## **Real-World Example**

During a session, your AI was supposed to trigger the **“Book Demo”** goal after detecting intent.  
Instead, it continued the conversation without completing it.

After clicking the **🤖 Debug icon**, you find:

**Condition “Intent = Demo Booking”** didn’t match because the user said “schedule a quick call” (not “demo”).

You update the condition to include synonyms (“book,” “call,” “schedule”).

You recreate the same session — and this time, the goal triggers perfectly.

That’s how Debugging turns insight into improvement — instantly.

## **Best Practices**

-   Debug sessions **after every major Journey or Brain update** to catch regressions early.
-   Always test multiple user variations of key phrases (“demo,” “meeting,” “talk”).
-   Keep Knowledge sources clean and well-tagged for accurate context retrieval.
-   Use **Agent Version labels** to organize iterative testing.
-   Link Debugging findings back to your **Playbooks** and **Guardrails**.

## **Why It’s Powerful**

Debugging doesn’t just fix problems — it builds intelligence.  
Each time you trace and correct your AI’s logic, you’re training it to think, respond, and sell smarter.

Your AI doesn’t learn from perfection — it learns from the fixes you make here.

> **Note:** Sessions cannot currently be deleted. This applies to both **“Test Bot” sessions** and **“Real Sessions.”** All sessions remain visible in the **Sessions** dashboard so you can review conversations and analyze how your AI Agent is performing.

## **Related Articles**

[Robot Head Icon](https://help.magicblocks.ai/en/article/troubleshooting-and-refining-your-ai-agent-using-the-robot-head-icon)

[Guardrails](https://help.magicblocks.ai/en/article/what-are-guardrails)

[Journey Advanced](https://help.magicblocks.ai/en/article/journey-advanced)