Advisor 1M checkbox - test
Pre-release
Adds the missing 1M checkbox beside Advisor model. It uses the same saved [1M] marker as the other models, preserves the selection when changing models, and clears when the advisor is disabled. Usage reports the canonical model name.
The advisor context byte guard is 2 MiB normally and 8 MiB when 1M is checked. The provider still determines its actual token limit; this checkbox cannot grant a larger context window to a model that lacks one.
Downloads: Windows ZIP includes the EXE and isolated-profile launcher; macos-15.zip is Apple Silicon and macos-15-intel.zip is Intel. Mac apps use full ad-hoc bundle signing and remain unnotarized; see README for Open Anyway instructions and the separate Mac launcher.
Source: 7079d00d4e206f75a76cfbb824a866c9f708ab29, branch feat/same-provider-advisor. This is a new test release; the previously tested advisor release is unchanged. No upstream advisor PR has been opened, pending manual Windows/Mac checks.
No new live model requests were made for this checkbox change. The original advisor's real-client proof is baseline evidence; the new 1M request wiring and size guards are covered by local HTTP/regression tests.
All three CI platforms passed 84 focused Rust tests and 24 UI tests, plus native builds. Both Mac apps passed strict signature verification before and after ZIP extraction, and native startup. TypeScript, formatting, and local Clippy also passed. Verification run.
Local Windows startup remains unverified for this revision because the older running test instance prevented a second copy from starting. Quit it before launching the update. Manual validation remains the gate for the upstream advisor PR.