# Inspector settings

> Look up what each panel of the Inspector's Settings page shows and the capture policy controls you can edit there.

The Inspector's Settings page: what each panel shows and which capture controls it edits. Open it with `blop open settings` (`/inspector?view=settings`). Options for browser checks are in [Configuration](/reference/configuration/).

## Settings overview panels

The overview is a read-only summary, not a connection test. Commands it shows run in your terminal and have **Copy** buttons; `--print` and hook previews change nothing. Secrets show only as `set` or `not set`, and secret copy chips copy an empty assignment. A lock means read-only on this page, not unchangeable.

| Panel | Shows | Fragment |
|---|---|---|
| **Connected** | MCP clients with a Blop entry, its registration state and config location. An entry is configuration, not proof of a session. | `#connected`, `#ov-settings-connected` |
| **Agents** | Coding agent binaries found, versions, hook support and setup commands. Detection is not proof of capture. | `#agents`, `#ov-settings-agents` |
| **Memory** | [Local memory](/guides/local-memory/) path, node count, nodes not updated in 90 days, index coverage. Unreadable shows unknown, not zero. | `#memory`, `#ov-settings-memory` |
| **Evidence** | Store path and size, retention hours and size cap. They come from the collector environment and need a restart. | `#evidence`, `#ov-settings-evidence` |
| **Ask model** | Chat provider, model, source and whether a key is set. Chat sends the evidence it reads to that provider. | `#ask`, `#ov-settings-ask` |

Detailed sections are `#settings-capture`, `#settings-storage`, `#settings-collector` and `#settings-usage`. Repair commands keep valid `--url` and `--port` pins; an unsafe or ambiguous pin hides the command instead of switching collectors.

## Capture freshness

Add a local project ID to see whether capture is arriving: `/inspector?view=settings&project=my-app#agents`. Without one, freshness is unknown.

| Rule | Value |
|---|---|
| Sample | The newest 1,000 collector observations from the last 7 days |
| Counted | Observations whose recorded hook project matches the ID |
| Recent | A match in the last 24 hours shows as capturing |
| No match | Does not prove capture was never set up: a busy collector can fill the sample with other activity |

## Capture and privacy

Under **Capture and privacy**, edit a binding's mode, body retention and excluded content classes, then choose **Save policy**. It uses the same writer as `blop agents capture-policy set`; see [What is captured](/guides/connect-agent/#what-is-captured).

| Mode | Content kept |
|---|---|
| `full` (default) | Prompts, public replies, tool input, output and errors, and other supported classes |
| `conversation-only` | Prompts and public model output |
| `metadata-only` | No bodies; lifecycle, model and usage metadata only |

| Control | Values |
|---|---|
| Exclusions | `prompt`, `model-output`, `tool-input`, `tool-output`, `tool-error`, `error`, `system`, `developer`, `compaction` |
| Body retention | Whole hours, 1–720; default 720 (30 days), measured from the original observation |
| Text caps | 16 KiB per text part, 64 KiB per event |

Invalid values are rejected before saving; a stale revision asks you to reload. Changes affect future events and pending delivery, never retained history, and editing a disabled binding does not enable it. Expired bodies show as retention markers; metadata follows the store's own retention (`BLOP_MONITOR_RETENTION_HOURS`, default 720, restart to apply), and size eviction can remove evidence earlier. Capture stays local and uploads nothing.