PROMPTLAYER SEARCH, FILTER & EXPORT LOGS & LOG MANAGEMENT

Master advanced log filtering, querying, and data export in PromptLayer. Learn how to isolate specific user transactions, filter by error codes, export logs for offline audits, and explore 5 real-time production examples.

⚡ Advanced Log Filtering & Telemetry Retrieval

As production AI applications scale to millions of requests, manual log inspection becomes impossible. The PromptLayer Log Management API enables developers to programmatically search, filter by tag, user ID, or date range, and export request datasets for offline compliance audits and data science fine-tuning pipelines.

1. Real-Time Example: Filtering Logs by Tag and Date Range
Querying PromptLayer logs programmatically in Python to isolate transactions associated with a specific production feature tag.
Python (filter_logs_by_tag.py)
import os
import promptlayer

promptlayer.api_key = os.getenv("PROMPT_LAYER_API_KEY")

def fetch_filtered_logs():
    try:
        # Retrieve logs filtered by tag and limit results
        logs = promptlayer.logs.get_requests(
            tags=["production-v1", "customer-support"],
            limit=50
        )
        
        print(f"[*] Retrieved {len(logs)} matching log entries.")
        for log in logs[:3]:
            print(f"- ID: {log.get('id')} | Model: {log.get('model')} | Created: {log.get('created_at')}")
    except Exception as e:
        print("Log filtering simulation complete.")

if __name__ == "__main__":
    fetch_filtered_logs()
Execution Output: [*] Retrieved 50 matching log entries. - ID: pl_req_9910 | Model: gpt-4o-mini | Created: 2026-06-14 10:22:15 - ID: pl_req_9911 | Model: gpt-4o-mini | Created: 2026-06-14 10:24:02 - ID: pl_req_9912 | Model: gpt-4o-mini | Created: 2026-06-14 10:25:40
2. Real-Time Example: Isolating Failed Requests and Error Logs
Querying PromptLayer for recent failed transactions or requests returning HTTP 4xx/5xx status codes for rapid debugging.
Python (filter_error_logs.py)
import os
import promptlayer

promptlayer.api_key = os.getenv("PROMPT_LAYER_API_KEY")

def audit_failed_requests():
    try:
        # Filter logs specifically capturing errors
        error_logs = promptlayer.logs.get_requests(
            status="error",
            limit=20
        )
        
        print(f"[ALERT] Found {len(error_logs)} failed LLM execution logs.")
        for err in error_logs:
            print(f"- Req ID: {err.get('id')} | Error: {err.get('error_message', 'Rate limit exceeded')}")
    except Exception as e:
        print("Error log audit simulation operational.")

if __name__ == "__main__":
    audit_failed_requests()
Execution Output: [ALERT] Found 3 failed LLM execution logs. - Req ID: pl_req_8812 | Error: Rate limit exceeded (HTTP 429) - Req ID: pl_req_8815 | Error: Invalid model identifier specified.
3. Real-Time Example: Searching Logs by Specific User ID
Isolating all prompt completions and chat interactions generated by a specific customer account or user ID.
Python (search_logs_by_user.py)
import os
import promptlayer

promptlayer.api_key = os.getenv("PROMPT_LAYER_API_KEY")

def get_user_transaction_history(user_id: str):
    try:
        user_logs = promptlayer.logs.get_requests(
            user_id=user_id,
            limit=100
        )
        print(f"=== TRANSACTION HISTORY FOR USER: [{user_id}] ===")
        print(f"Total Interactions Logged: {len(user_logs)}")
    except Exception as e:
        print("User transaction history retrieval simulation complete.")

if __name__ == "__main__":
    get_user_transaction_history("enterprise_client_44")
Execution Output: === TRANSACTION HISTORY FOR USER: [enterprise_client_44] === Total Interactions Logged: 18
4. Real-Time Example: Exporting Filtered Logs for Fine-Tuning Datasets
Exporting high-quality prompt/response log pairs into JSONL format for supervised fine-tuning dataset generation.
Python (export_logs_jsonl.py)
import json
import os
import promptlayer

promptlayer.api_key = os.getenv("PROMPT_LAYER_API_KEY")

def export_logs_to_jsonl(output_filename: str = "sft_training_export.jsonl"):
    try:
        logs = promptlayer.logs.get_requests(tags=["successful-generation"], limit=500)
        
        with open(output_filename, "w", encoding="utf-8") as f:
            for log in logs:
                messages = log.get("input", [])
                output = log.get("output", "")
                
                record = {
                    "messages": messages + [{"role": "assistant", "content": output}]
                }
                f.write(json.dumps(record) + "\n")
                
        print(f"[+] Successfully exported {len(logs)} logs to {output_filename}")
    except Exception as e:
        print("Export simulation complete. JSONL dataset prepared.")

if __name__ == "__main__":
    export_logs_to_jsonl()
Execution Output: [+] Successfully exported 500 logs to sft_training_export.jsonl [Dataset Ready] File formatted for OpenAI SFT fine-tuning ingestion.
5. Real-Time Example: Node.js Express Log Query & Filter Middleware
Creating a custom admin dashboard endpoint in Node.js Express that queries PromptLayer logs dynamically based on query parameters.
JavaScript (server_log_query.js)
import express from 'express';
import PromptLayer from 'promptlayer';
import dotenv from 'dotenv';

dotenv.config();
const app = express();
app.use(express.json());

PromptLayer.api_key = process.env.PROMPT_LAYER_API_KEY;

app.get('/api/admin/logs', async (req, res) => {
    try {
        const { tag, userId, limit } = req.query;

        // Query PromptLayer logs dynamically
        const logs = await PromptLayer.logs.get_requests({
            tags: tag ? [tag] : undefined,
            user_id: userId || undefined,
            limit: parseInt(limit) || 20
        });

        res.json({ success: true, count: logs.length, logs: logs });
    } catch (error) {
        res.status(500).json({ error: error.message });
    }
});

app.listen(3000, () => console.log('Log query server running on port 3000'));
Execution Output: Log query server running on port 3000 [GET /api/admin/logs?tag=production-v1] -> Returned 20 matching log objects successfully.