OpenAI-compatible BYOK · Skills marketplace · Safer sharing

Model connections

• You can now connect Anthropic-compatible or OpenAI-compatible credentials, choose Chat Completions or Responses, and use the configured model in Tasks and Timers without platform model charges.
• Provider compatibility now preserves Anthropic API versions and Responses conversation history, while deleting a Timer releases its credential reference cleanly.

Skills marketplace

• Featured skill collections are easier to browse with explicit ordering, richer descriptions, stable detail links, and installed cards that remain visible.
• GitHub skill imports now handle redirected and single-skill repositories more reliably, while marketplace and settings copy follows your selected language.

Web research and articles

• Known web links now use managed web search and safe prefetching more consistently, and imported articles preserve link-rich content.

Sharing and previews

• Shared Markdown images render more safely, share previews and skill uploads recover correctly, and link-reference actions no longer appear where they cannot be used.

Bots and Timers

• WeChat Bot conversations are isolated by Bot so separate assistants no longer reuse the same chat binding.
• Timers using a personal model credential can be removed without leaving a reference that blocks credential cleanup.

Signup and account protection

• Email signup now uses a direct image challenge, with stronger abuse controls for rapid signup clusters and reward eligibility.
• Google sign-in now carries callback state consistently through Casdoor and desktop handoff, preventing empty-state callback failures.

Settings and composer

• Settings failures are contained instead of breaking the whole page, and the composer shows clearer skill descriptions with a more compact trigger.

Updates

• The Updates page now uses a more consistent visual palette.