LLM Logs

Trace model calls, token use, cost, and latency.

LLM Logs records every LLM completion and embedding call the platform made. Open Control Center → Observe → LLM Logs, on either the Gateway Logs or Embedding Logs tab.

Gateway Logs

Gateway Logs records every completion call, with its model, target, tokens, and cost.

Gateway Logs, tracing every completed model call.

You can narrow by project and model. You can also search by model or agent. Each row is one completed model call:

Field	What it shows
Model	The model that ran, such as gpt-4.1-mini.
For	The target the call was made for.
Tokens	Tokens consumed by the call.
Cost	The credit cost of the call.
Latency	How long the call took, in milliseconds.
When	The completion time.
Status	The call outcome, such as Success.

Embedding Logs

Every query sent to an embedding model gets an entry in Embedding Logs, from any provider and any part of the platform, including calls that fail. It records embedding calls, not completions.

Embedding Logs, with one entry expanded.

You can narrow by project, model, and date range. You can also search by query or model. Each row is one embedding call:

Field	What it shows
Query	The text that was embedded — a document chunk, a search query, or an instruction.
Model	The embedding model that ran, such as OpenAI Text Embedding Ada 002.
When	The completion time.
Status	Success or Failed.

Select a row on either tab to see the call's full detail. On Embedding Logs, an expanded row shows the full query text, the model, and how long the call took, in milliseconds.

---

Maintained by Abhishek Paul AI-assisted, human-approved