Migrate from freedictionaryapi.com

If your app uses freedictionaryapi.com, change the base URL to a Vocab Bloom Hub instance and keep the English lookup path. The compatibility adapter returns the familiar { word, entries, source } response, with additive vocabBloom source details. Its answers come from the active dataset of your instance. The upstream service uses Wiktionary data by default; the project instance serves its own dataset by default.

Change the base URL

For example:

curl -fsS 'https://vocab-bloom-hub.com/api/compat/freedictionaryapi/v1/entries/en/hello?translations=true'
curl -fsS 'https://vocab-bloom-hub.com/api/compat/freedictionaryapi/v1/languages?pretty=true'

These are anonymous, read-only requests. They do not contact freedictionaryapi.com. translations and pretty accept lowercase true or false; translations are omitted by default. The adapter treats spelling as case-sensitive and returns 200 with empty entries for a missing word. It accepts en and all for the English database; other language codes return empty entries. /languages lists English only, regardless of translation languages inside entries.

The adapter does not accept ?dataset=.... To serve a different dataset in this response shape, install it on a PostgreSQL instance and activate it in the admin UI. Activation changes the dataset of all public lookups on that instance. SQLite has only the default dataset.

Read every dataset in this format

Append /datasets to the compatibility lookup. translations and pretty work here too:

curl -fsS 'http://localhost:3240/api/compat/freedictionaryapi/v1/entries/en/hello/datasets?translations=true&pretty=true'

This Vocab Bloom Hub extension reads every installed dataset without activating one. Its outer response is { data, meta }. Each data[] group contains the dataset terms (including dataset, title, active, source and licenses), status: 200, and a response containing that dataset's full { word, entries, source, vocabBloom } adapter object.

Spelling remains case-sensitive in each dataset. A missing word or an unsupported language has empty response.entries, as in the ordinary adapter. The overall HTTP status is 200 even if every group is empty. meta contains the requested word and language, the number of datasets, and found (datasets with nonempty entries).

For example, extract the OpenGloss response:

curl -fsS 'http://localhost:3240/api/compat/freedictionaryapi/v1/entries/en/hello/datasets?translations=true' \
  | jq '.data[] | select(.dataset == "opengloss") | .response'

Only installed datasets appear. SQLite returns the default dataset only. Terms always identify the dataset read; an empty result links back to this all-datasets route rather than the active-only /api/v1/meta. The ordinary lookup URL keeps its original response shape and reads the active dataset. See the complete all-datasets contract.

Check your client before switching

• The adapter preserves word, entries, source, senses, forms and supported translations, but individual senses, forms, tags and counts depend on the installed data.
• Keep vocabBloom and its license/origin details when storing or redistributing results. A single source.license cannot express every source and contribution term.
• Upstream supports more lookup languages. This instance currently stores English headwords; its translations are words inside English entries, not separate language dictionaries.
• The instance has its own rate limit and cache policy. Check representative words, casing, missing-word handling and translations=true before switching production traffic.

For the complete mapping and HTTP limits, see freedictionaryapi.com compatibility. For dataset installation and activation, see Datasets.

Website comparison snapshots

The website comparison uses saved original-service responses for all nine preset words and supported version/translation options. They are committed with the website and refreshed manually with yarn workspace site comparisons:generate (--missing-only retries missing or failed snapshots). Building, starting and browsing the site never requests the originals. The retrieval date is displayed. A failed refresh retains an older successful snapshot with a warning when available; otherwise the error is shown. Instance datasets are queried live.