I loaded 30 days of real LLM traces into a live demo. Here is what they reveal
Description and use
The Torrix live demo lets users instantly explore real LLM traces, spot cost spikes, compare model costs, and view detailed agent traces—all without installation.
Problem
Developers struggle to understand LLM usage costs, anomalous spikes, and model performance without intrusive self‑hosted tools.
Outcome
The live demo demonstrates how real LLM traces can reveal cost spikes, hidden expensive models, detailed agent execution, and evaluation performance, all without installation. It lowers the barrier to try Torrix and encourages adoption.
Workflow
- 1. Open the live demo URL (demo.torrix.ai).
- 2. Browse the Analytics tab to see cost spikes and model cost breakdowns.
- 3. Click on a flagged anomaly to view prompt, model, and token count.
- 4. Inspect the 5‑step agent trace view for detailed step logs and timings.
- 5. Run evaluation results for test datasets and review pass rates.
- 6. Execute custom SQL queries against the underlying SQLite database.
- 7. Export data to CSV or use the schema browser.
- 8. (Optional) Self‑host by running a single Docker container with torrixai/torrix:latest.
Evidence
The demo loads 30 days of LLM traces across 3 simulated projects: Production API, Data Pipeline, Customer Support Bot.
On days 14 and 15, call volume tripled overnight — 55 requests per day vs the normal 18. Every anomalous run is flagged with a SPIKE badge automatically.
claude-3-5-sonnet handles 35% of traffic at $3.00/$15.00 per million tokens and drives the majority of spend.
- Record date: 2026-09-06T23:37:23.620Z