durindoor-embeddings

Files (1)

• SKILL.md

SKILL.md (first lines)

# DurinDoor Embeddings

## Discover

```bash
curl -H "Authorization: Bearer $DURINDOOR_KEY" "$DURINDOOR_URL/v1/models/embedding" | jq -r '.data[].id'
MODEL_ID="$(curl -s -H "Authorization: Bearer $DURINDOOR_KEY" "$DURINDOOR_URL/v1/models/embedding" | jq -r '.data[0].id')"
curl -H "Authorization: Bearer $DURINDOOR_KEY" "$DURINDOOR_URL/v1/models/info?id=$MODEL_ID"
```

## Embed text

```bash
curl -X POST "$DURINDOOR_URL/v1/embeddings" \
  -H "Authorization: Bearer $DURINDOOR_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"$MODEL_ID\",\"input\":[\"hello\",\"world\"]}"
```

`input` accepts a string or array. Optional dimensions, encoding format, and batch limits depend on the selected model. The response uses OpenAI-compatible `data[].embedding` arrays.

## Use the default route

Leave out `model` and DurinDoor uses the first available model in the embeddings route (**Dashboard → Media Routes**). Embeddings never switch to a second model, because vectors from different models are not comparable. Pin `model` when you store vectors, so later calls use the same model.

```bash
curl -X POST "$DURINDOOR_URL/v1/embeddings" \
  -H "Authorization: Bearer $DURINDOOR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input":"hello"}'
```

If no connected provider serves this endpoint, the response is HTTP 400 with `error.code` set to `no_provider_for_kind`. Connect a provider, or pass `model`.

Reference: https://github.com/bloodf/durindoor/blob/main/docs/reference/api.mdx

From bloodf/durindoor, MIT.

More from bloodf/durindoor