Helpdock 1.0.0

The first stable release. Every milestone of the 1.0 plan has shipped: the
ticket desk with SLAs, rules and macros; email, Telegram, WhatsApp, Messenger
and Instagram, the chat widget and the web form; a bilingual help center on
each brand's own domain; grounded AI that assists agents and answers customers
from the brand's knowledge; the REST API, webhooks and reports; staff sign-in
over SAML and OpenID Connect; and importers from Zendesk, Freshdesk and Zoho
Desk with a manual CSV mapper for the rest. What each milestone built, and the
gaps it left open, is in docs/completed/.

New since 0.7

• Previous tickets on the contact card. The ticket's details panel lists
  up to three of the contact's other tickets, newest first, with their status,
  a count of every earlier ticket you may see, and a link to the contact's
  page (#239).
• Auto-reply answers only what the sources answer. When the knowledge base
  does not answer the customer's exact question, the model replies
  NO_ANSWER and the conversation goes to the team, instead of offering
  related facts. The CONFIDENCE: figure is read on its own line or run on
  from the last sentence, in any case or emphasis
  (#238).
• "AI answered" on the ticket list appears only once the assistant has
  sent an answer (#233).

Upgrading from 0.7

Nothing to change in your configuration. The major version marks the release,
not an incompatible change: follow
Upgrading with 1.0.0, and the api runs
any pending migrations when it starts.

Before the release

• Release gates. The external pentest of the widget and the API, the load
  run on the reference host, the screen-reader pass, the onboarding run on a
  clean VM with three outside testers, and the restore drill were completed
  before 1.0.0. Their results are held by the maintainer and are not published
  in this repository.
• AI answer quality. The bilingual evaluation (DOMAIN-RULES §9) passed
  every threshold in run 39322685646,
  with glm-4.6 answering, glm-5.1 judging and bge-m3:567m embeddings:
  answerable questions answered correctly 97.5 % in English and 94.6 % in
  Arabic, unanswerable ones handed off 100 % in both, no injected instruction
  followed, every citation resolved, every answer in the question's language.
  That is one provider pair; evaluate the model you choose
  (AI guide).
• OWASP ASVS Level 2. Every requirement has been walked with its evidence,
  and no gap is open (asvs-l2).

Known limits

• SSO and imports were not run against live accounts. Sign-in was verified
  against stand-in OIDC and SAML identity providers, and the Zendesk, Freshdesk
  and Zoho Desk importers against stand-in servers built from each vendor's
  published API documentation. No run against a live Okta or Entra ID tenant
  or a live vendor account was made before 1.0.
• More importers come later. Help Scout, Intercom, Gorgias, Zammad and
  Front connectors are deferred; until then their exports go through the manual
  CSV mapper (importing).

What's Changed

• docs: settle the empty Telegram welcome on the default text by @Docker-Hunterpedia in #232
• fix: say "AI answered" only once the assistant has answered by @Docker-Hunterpedia in #233
• ci: run the AI evaluation with a chat-only key and local embeddings by @Docker-Hunterpedia in #234
• fix: hand off what the knowledge base does not answer, and read the confidence trailer reliably by @Docker-Hunterpedia in #238
• feat: previous tickets on the ticket's contact card by @Docker-Hunterpedia in #239
• chore: prepare Helpdock 1.0.0 by @Docker-Hunterpedia in #235
• chore: version packages by @github-actions[bot] in #236

Full Changelog: v0.7.0...v1.0.0