SeekChatZot 1.0.0rc8

• Fix: tool calls failed with servers that repeat the tool name in every stream chunk (e.g. qwen3 via Ollama):
  the name was appended each time (search_librarysearch_library…), so the tool was "unknown" and every retry failed.
  A repeated name now replaces instead of appends.
• Requests carry User-Agent: SeekChatZot/<version>, so the server log shows who asked.

SeekChatZot 1.0.0rc7

• Fix: the user name now reaches LiteLLM. It is sent in the OpenAI-standard user field of the chat body (rc.6
  used end_user, which the proxy ignores); the statistics show it as end_user.
• Settings: checkbox rows ("think before answering", "send my user name") use the full width, the checkbox sits next
  to its text.

SeekChatZot 1.0.0rc6

• User name for the usage statistics: with ollama.ils.local as endpoint, SeekChatZot sends the Zotero account
  name as end_user in the chat body. New setting "Send my user name …" (default on); off: only the SHA-256 hash of
  the name is sent. The setting is disabled for any other endpoint, where nothing is sent.

SeekChatZot 1.0.0rc5

• Renamed to SeekChatZot (the name SeekChat is taken): name in Zotero, menus, window titles, settings, export
  headings and the XPI file (seekchatzot-<version>.xpi). Add-on id seekchat@ils-forschung.de, the settings and
  the update address stay, so installed copies update and keep their settings; the repositories keep their names.

SeekChat 1.0.0rc4

• New name: the tool chat is now called "Chat with your Literature" / „Chatte mit deiner Literatur“ (button,
  window title, menu, export heading).
• Tools menu: "SeekChat: Chat with your Literature" opens it from Zotero's Tools menu, in every Zotero version.
• Library chat disabled: no toolbar button next to ZotSeek's, no "Chat with this selection" in the item context
  menu, no menu entry. The code is commented out, not removed; the window itself still works and stays tested.

SeekChat 1.0.0rc3

• Server error 5xx: one more try. A request that the model server answers with HTTP 5xx before anything was
  streamed is sent once more after a second. Ollama answers 500 when it cannot parse a tool call the model wrote
  ("XML syntax error … unexpected EOF", seen in the live test with a subagent); the second sample usually works.

SeekChat 1.0.0rc2

• Only the OpenAI-compatible interface: chat, model list and tools go through /v1/models and
  /v1/chat/completions for every server, so vLLM (and llama.cpp, LM Studio …) work like Ollama. The setting
  "Interface" is gone: the server URL is the root of an Ollama server (SeekChat adds /v1) or the full /v1 base.
• Context window: taken from /v1/models (vLLM max_model_len); Ollama does not report it there, so SeekChat asks
  its native API read-only (/api/ps, /api/show) – the only Ollama-specific call, ignored on other servers.
  SeekChat still never sends a context size.
• Thinking switch in the OpenAI form: reasoning_effort: "none" (Ollama) and
  chat_template_kwargs.enable_thinking (vLLM/SGLang); a server that refuses them is asked without them.
• README: the outdated note on /api/chat and num_ctx is corrected.

SeekChat 1.0.0rc1

• Certificates: when a chat server is entered in the settings (server URL or allowed hosts changed, or "Check
  certificate" clicked), SeekChat checks its https certificate. If the system does not trust it, a dialog shows the
  certificate (subject, issuer, validity, SHA-256, reason; "Show certificate" with the PEM) and asks: trust or not
  trust. The decisions and the certificates themselves (DER) are listed under "Certificates" in the settings, where each
  can be switched between trusted and not trusted, shown, copied as PEM or removed.
• Only a trusted certificate with exactly this fingerprint is accepted for its host and port (pinning); a new
  certificate on the same server is refused until it is decided. Before, "Accept invalid certificate" accepted
  whatever certificate the server showed first.
• The old setting "Accept invalid certificate" is gone: who had it on gets the server's current certificate entered as
  trusted once (visible in the list).
• Homepage and updates on gitlab.com: the plugin's metadata link to
  https://gitlab.com/ils-research/zotero-plugins/seekchat-zotero, and updates now come from the releases there
  (update_url → updates.json in that repository, release by the CI on the tag). The portal still
  offers the download and updates installations from before this version.
• Pre-releases: package.json holds semver (1.0.0-rc.1), the XPI Mozilla's form (1.0.0rc1, before 1.0.0);
  scripts/publish.py sorts them so. E2E: the mock also serves https with a self-signed certificate;
  ZOTERO_VERSION=… ./e2e/run.sh picks the Zotero version.