Logs & Monitoring

This article covers LanSphere’s runtime observability for apps: review run records and annotate in the “Conversation Logs” tab, track overall metrics in the “Data Dashboard” tab, and export logs for archiving and analysis. It is for developers, admins, and owners responsible for app operations and quality improvement.

After an app goes live, continuously observing its runtime is key to service quality. We recommend forming a daily operations loop with three tools: spot trends and anomalies on the Data Dashboard, locate concrete problems in Conversation Logs, and accumulate optimization material through annotation.

Conversation Logs

The “Conversation Logs” tab on the app details page shows run records by app type — Workflow logs and conversational-app logs are structured differently.

Tip: The “Conversation Logs” tab carries both log viewing and annotation — think of it as “conversation logs and annotation.” Agent apps have the same tab and use it the same way.

Workflow Logs

A Workflow is a single-run automation app; logs are recorded per run, each entry with these fields:

Chatflow and Agent Logs

Conversational apps organize logs by conversation:

• Each conversation shows how many messages it contains;
• Each message records input, output, elapsed time, and Token consumption;
• User feedback on answers (Like / Dislike) is also shown.

Viewing a Single Run’s Details

Click a log entry for the full run details:

• Workflow: Inspect input, output, and intermediate data node by node, reconstructing the full execution chain;
• Conversational apps: Read message by message with conversation context, reconstructing the interaction.

When an app misbehaves or an answer misses expectations, find the problem log by time or status first, then use the single-run details to locate step by step which node or which message went wrong — and go back to the “App Development” tab to adjust orchestration, prompts, or Knowledge Base configuration accordingly.

Annotation

In Conversation Logs, you can annotate answers to keep improving answer quality:

• Like / Dislike: Mark answer quality to quickly separate good answers from problem answers;
• Annotated reply: For an unsatisfying answer, add the reply you expected — recording the right way to respond.

Example scenarios:

1. Periodically screen topics with many Dislikes and judge whether the problem is missing Knowledge Base content, insufficient prompt constraints, or poor orchestration logic;
2. Add annotated replies to typical problem answers as references for later optimization;
3. Feed accumulated annotations into refining prompts, supplementing Knowledge Base content, or adjusting orchestration logic — forming a “find problems — annotate — improve the app” loop.

Annotation is long-term work. Build it into the app’s daily operations: assign someone to process Dislike records regularly, and after each optimization, look back and verify whether answers to the same kind of problem have improved.

Data Dashboard

The “Data Dashboard” tab shows the app’s overall runtime in visual charts. Main metrics:

Charts can be filtered by time range. Compare metric changes across periods: when the success rate drops suddenly, narrow the range to locate where the anomaly started, then find the failure records for that period in Conversation Logs; when Token usage grows abnormally, check the logs for abnormal calls.

Log Export

When you need to archive logs or analyze them in depth off-platform, export them as a JSON file.

Permissions

Only Owners and Admins see the “Export” button; other roles cannot export logs.

Steps

1. Open the app’s “Conversation Logs” tab and click “Export”;
2. Pick a time range in the pop-up — nine options:

3. Confirm to export a JSON file named like conversation_log_20260729_120000.json (ConversationLog_ExportDate_ExportTime.json).

Tip: If there is no data in the selected range, the system prompts “No matching logs” — adjust the range and retry.

About the Exported File

The exported content’s fields match what is shown in the “Conversation Logs” tab, organized by app type:

• Workflow apps: Each run’s status, start time, elapsed time, total Tokens, input/output, and node intermediate data;
• Chatflow and Agent apps: Conversation-level message counts, plus per-message input, output, elapsed time, Tokens, and user feedback.

User feedback is recorded as codes:

Related Reading

• For publishing methods and channel configuration, see “Publishing Channels”;
• For app permissions, ownership transfer, and deletion rules, see “App Management”;
• For orchestrating and debugging Chatflow and Agent apps, see “Chatflow” and “Agent.”