Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion memory-bank/aiConfig.md
Original file line number Diff line number Diff line change
Expand Up @@ -649,7 +649,8 @@ known generation, the existing GPT-5.x and GPT-6 catch-alls still apply.
Gemini (`gemini-helpers.ts`) keys its entries by a representative id per
variant and version (`gemini-3.7-flash`, `gemini-3-pro`); the exported
`findGeminiVariantDonor` (`gemini-variant.ts`) parses
`gemini-<major>[.<minor>]-<pro|flash|flash-lite>` and returns the
`gemini-<major>[.<minor>]-<variant>` (`pro`, `flash`, `flash-lite`, and the
Gemini 4 names `argon`, `carbon`, `hydrogen`, `neon`) and returns the
same-variant known id with the greatest version at or below the requested
one, plus whether it is the variant's newest: `gemini-3.9-flash` → 3.7/3.8
Flash, `gemini-4-pro` → 3 Pro. The table ignores the newest flag; the
Expand Down
8 changes: 5 additions & 3 deletions memory-bank/geminiInteractions.md
Original file line number Diff line number Diff line change
Expand Up @@ -124,9 +124,11 @@ Gemini 3.6/3.7 Flash were invisible until manually allowlisted):
- **Future versions inherit.** `getGeminiInteractionsProfile` falls back,
after an exact miss, to ai-config's `findGeminiVariantDonor`: a Gemini
chat id (`isGeminiApiChatModel`) whose variant (`pro`, `flash`,
`flash-lite`) version is at or above that variant's newest profiled id
takes that profile (`gemini-3.9-flash` → `gemini-3.8-flash`,
`gemini-4-pro` and the alias `gemini-3.1-pro` → `gemini-3.1-pro-preview`).
`flash-lite`, or a Gemini 4 name: `argon`, `carbon`, `hydrogen`, `neon`)
version is at or above that variant's newest profiled id takes that
profile (`gemini-3.9-flash` → `gemini-3.8-flash`, `gemini-4-pro` and the
alias `gemini-3.1-pro` → `gemini-3.1-pro-preview`,
`gemini-4-argon-cyber-permissive` → `gemini-4-argon`).
An id below the newest profile (`gemini-3-pro-preview`) gets nothing,
since plain "nearest older" would hand it 2.5 Pro's `medium`. The chat gate
matters because `GeminiClient` looks up whatever id it is given:
Expand Down
3 changes: 3 additions & 0 deletions memory-bank/opencodeRouting.md
Original file line number Diff line number Diff line change
Expand Up @@ -118,6 +118,9 @@ by the resolved credential base URL so a product switch refetches.
`thinkingConfig`; the 3.x rules are exact per documented variant (level sets differ per variant —
3.7/3.8 Flash accept `low`/`medium`/`high`, no `minimal`; verified against
https://ai.google.dev/gemini-api/docs/thinking, page updated 2026-09-09 UTC, reviewed 2026-09-11).
Gemini 4 Argon and Carbon accept `low`/`medium`/`high`; Hydrogen and Neon add `xhigh` (reviewed
2026-11-20). `@ai-sdk/google` before 4.0.119 rejects `xhigh` in provider options without sending a
request, so the SDK must not drop below that version.
OpenCode discovery **excludes** a Gemini-stamped model with no verified profile (actionable
`logger.warn` diagnostic) rather than advertising an unusable model; an explicit configured model
with such an id surfaces the client's unsupported-variant error. No silent Chat Completions
Expand Down
192 changes: 91 additions & 101 deletions package-lock.json

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

Loading
Loading