Semantic search
The Semantic search tab of the System screen. Search by meaning finds a record by what it is about rather than by the exact words in it. This tab decides where that meaning is computed.
Search by meaning is used by the documentation, tasks, Drive, creatives and the assistant. Turning it off does not leave those screens without search: the plain word search keeps working wherever it exists.
Where vectors are computed
The Search mode switch has three positions:
- External (OpenRouter) — the text is sent to OpenRouter and billed to your key there.
- Internal (on this box) — the text is processed by an engine running on this server and never leaves it. This is the position a freshly installed box starts on.
- Off — search by meaning is off everywhere.
Cores and memory for the local engine
Both settings apply to the Internal (on this box) position and take effect after you press Save next to the field.
- Embedding service cores — how many processor cores the local engine may use: a whole number, at least 1. More cores mean faster vectors and less headroom for everything else on the server. The default is 4.
- Embedding service memory — how much memory it may use, in gigabytes, from 0.5 to 256. Too little and the engine will not fit and gets restarted; too much leaves less for the rest of the box. The default is 3 GB.
Model
The Embedding model block is read-only: it shows the model actually in use and the length of the vectors it produces. There is nothing to choose here — the box runs one model.
Under it is a line about the current state:
| The line says | What it means |
|---|---|
| Semantic search is active. | The cloud position is set up and working. |
| External mode is selected, but this box has no OpenRouter key configured — semantic search stays unavailable until the key is set in the Connections section. | The cloud position is selected, but the box has no OpenRouter key. Add it in Connections. |
| Semantic search is active — vectors are computed locally on this box, nothing is sent to OpenRouter. | The local engine is in place and nothing leaves the server. |
| Internal mode is selected, but the local engine is not installed on this server yet, so semantic search is unavailable (word search keeps working). The engine is about twelve gigabytes and downloads after the mode is chosen. | The local position is selected, but the engine is not on this server yet. |
| Semantic search is off — it is disabled for documentation, tasks, Drive, creatives and the assistant. Word search keeps working where it is available. | The switch is in the Off position. |
The engine is downloaded to the server after the local position is chosen, and it is a large download — on a new box that is the state you will meet first. Picking the same position again asks for it once more, which is how you retry a download that failed. While you stay on the tab it re-checks on its own, so an engine that arrives soon is picked up without a page reload. That watch is bounded, though — on a slow link the download can outlast it; if the tab still reports the engine as missing, reload the page to start a fresh wait.
This is the same engine the Speech recognition tab uses, so once it has arrived for one of the two tabs, it is in place for both.