> ## Documentation Index
> Fetch the complete documentation index at: https://docs.memorylake.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Observability

> Trace every request, read the console call logs, and understand the error contract

Every request through Memory Router is traceable end to end — from your client, through the gateway, to the memory layer and the model.

## Request Tracing

Each response carries an **`X-Trace-ID`** header — a W3C-compatible trace id that spans the whole call chain. Log it alongside your own request ids; when something looks off, it is the fastest way for support to locate the exact call.

Memory Router also honors inbound **W3C Trace Context** (`traceparent` / `tracestate`) headers, so if your application already propagates distributed traces, the Router's spans join your existing trace.

## Console Call Logs

Every call lands in the console **Logs** page with per-call details:

| Field             | What it tells you                                                   |
| ----------------- | ------------------------------------------------------------------- |
| **Model**         | The canonical model that served the request.                        |
| **Tokens**        | Prompt, completion, and total token counts as metered by the model. |
| **Finish reason** | Why generation stopped (`stop`, `length`, tool call, …).            |
| **Key source**    | `byok` (your provider key) or `platform` (MemoryLake-hosted).       |
| **Trace ID**      | Matches the `X-Trace-ID` response header for cross-referencing.     |

<Tip>
  Use **Key source** to verify BYOK is actually in effect, and compare token counts across turns to see the effect of memory injection versus replaying full history.
</Tip>

## Error Contract

| Status                        | Meaning                                                              | What to do                                                                         |
| ----------------------------- | -------------------------------------------------------------------- | ---------------------------------------------------------------------------------- |
| `401 Unauthorized`            | Missing or invalid MemoryLake key.                                   | Check the key in the native auth header (hosted) or `x-memorylake-api-key` (BYOK). |
| `402 Payment Required`        | Account balance below the minimum required to forward the call.      | Top up your balance in the console, then retry.                                    |
| `502 Bad Gateway`             | The memory gateway could not reach its upstream service.             | Retry with backoff; if it persists, contact support with your `X-Trace-ID`.        |
| Provider errors (`4xx`/`5xx`) | Returned by the model provider itself, in the provider's own format. | Handle exactly as you would on a direct provider call.                             |

<Note>
  Requests **without** a `boundary_id` skip the memory layer but still authenticate and meter through the gateway. If you want to isolate whether an issue is memory-related, retry the same call without `boundary_id` and compare.
</Note>

## Next Steps

<CardGroup cols={2}>
  <Card title="Deployment Modes" icon="key" href="/features/memory-router/deployment-modes">
    BYOK vs. hosted, endpoints, and supported providers.
  </Card>

  <Card title="FAQ" icon="circle-help" href="/features/memory-router/faq">
    Common questions about code, providers, security, and availability.
  </Card>
</CardGroup>