Zactonz AI Connector for DeepSeek

1.0.2 — 29 October 2026#

Fixed

• A reply arrived empty when the provider answered with a list of content chunks rather than a single string, which the AI Client's parser does not read. The chunks were dropped, so the request looked successful and returned nothing.
• Models the provider serves for other tasks, such as transcription, speech and moderation, were offered as text models. The task the provider declares is now believed rather than guessed from the model's name.
• Entries the provider marks as internal reached the model picker. They answer the listing but are not products.
• The reasoning effort selector could offer a level the connector would not then accept. The list the selector is built from and the list that is accepted are now the same one, and are tested against each other.

Changed

• The reasoning effort description said a model that always reasons ignores the setting. Which levels a model accepts is the provider's business, and a model may ignore or refuse the level chosen.
• The base URL override wording now says that WordPress restricts which hosts and ports an outbound request may reach.

1.0.1 — 6 October 2026#

Fixed

• The settings screen reported no saved API key, and hid the control that removes it, on a site where AI support is switched off. The key is read back from the AI Client, which WordPress only populates while it is wiring connectors.
• The model-discovery answer in the FAQ described capability detection for embeddings, which this connector does not offer, and claimed capabilities are read from the provider rather than guessed. Where DeepSeek publishes no capability data for a model the connector falls back to inference.
• A streaming request whose parameters could not be encoded as JSON, malformed UTF-8 in post content being the realistic case, sent an empty body and drew a puzzling error from the provider. It now reports what actually went wrong.

1.0.0 — 30 September 2026#

• Initial release.
• Registers DeepSeek with the WordPress AI Client and the Settings › Connectors screen.
• Discovers models from the DeepSeek catalogue, with display names, context windows and accepted input types, so vision is offered only where DeepSeek accepts images.
• JSON output rather than a JSON schema, matching what DeepSeek accepts.
• Thinking-mode control for the default text model, with thinking output recorded separately from the answer.
• Cancellable server-sent event streaming for content, thinking and tool-call chunks.
• Redacted diagnostics report and an asynchronous WordPress Site Health connectivity test.
• API key from a PHP constant or an environment variable.