mnfst/manifest manifest@5.24.0

v5.24.0

on GitHub

✨ Minor Changes

• 4e08bb4: Add OAuth/subscription routing support for Anthropic Claude tokens

  • Subscription tab now accepts Claude setup-tokens (sk-ant-oat) with dedicated input UI
  • Backend stores and proxies subscription tokens (previously rejected)
  • Proxy sends correct Authorization: Bearer + anthropic-beta headers for subscription tokens
  • Fix case-insensitive provider matching for subscription cost/auth_type inference
  • Fix DELETE provider endpoint rejecting requests with validation error
  • Fix token whitespace corruption when pasting from terminal
  • Subscription badge overlay on provider icons in message log and overview
  • Proxy messages now store auth_type and set cost to zero for subscriptions
  • Fix duplicate messages: OTLP dedup remaps trace to proxy-recorded message
  • Conditional rollup preserves proxy token data instead of overwriting
  • ModelPickerModal always shows subscription/API key tabs with contextual empty states
  • Purge non-curated models after OpenRouter sync in local mode
  • Tier auto-assign excludes OpenRouter models from prefix-based provider inference

• dab83ca: Replace 3-mode system (cloud/local/dev) with 2-axis configuration (mode × devMode). The mode field now only accepts cloud or local for deployment model, while devMode (boolean) independently controls development behaviors like skipping API keys and faster metrics. The old mode: "dev" is still accepted for backward compatibility but logs a deprecation warning. devMode is auto-detected when the endpoint is a loopback address and no mnfst_* API key is provided.

🐛 Patch Changes

• 8758f8c: Add auth badge (subscription/api key) to provider icons across all pages for consistency
• Fix auth_type propagation on error/fallback records, dual-auth subscription billing, migration index names, and SSE passthrough buffer overflow
• 3d4eb56: Reduce custom-providers endpoint latency with caching and query parallelization
• 100b33d: Improve frontend performance with route-level code splitting, vendor chunk separation, and asset loading optimizations
• 3efa632: Optimize slow backend endpoints for sub-500ms response times: parallelize independent DB queries, batch saves, pre-fetch dedup context, derive summaries from timeseries data, and group notification cron evaluation.