mirror of
https://github.com/stack-auth/stack.git
synced 2026-07-20 21:29:36 +08:00
MantraandGitHub
81068977ff
## What
Refresh the AI model selection matrix and fix a few issues in custom
dashboard generation.
### Models (`apps/backend/src/lib/ai/models.ts`)
- Replace deprecated/placeholder model IDs with current ones:
- `smart/slow` authenticated → `openai/gpt-5.5` (was
`x-ai/grok-build-0.1`)
- `smart/fast` → `google/gemini-3.5-flash`
- `smartest` unauthenticated tiers → `z-ai/glm-5.2` /
`google/gemini-3.5-flash` (was `deepseek/deepseek-v4-flash`)
- `dumb` unauthenticated tiers → `nvidia/nemotron-3-super-120b-a12b`
### Email template rewrite
- Forward `x-stack-*` / `x-hexclave-*` headers from the caller through
the template-source rewrite route so the inner AI call
(`/ai/query/generate`) is authenticated and resolves to the
**authenticated** model tier instead of falling back to the
unauthenticated one.
- Lower rewrite quality to `dumb` / `slow` (sufficient for this task,
cheaper/faster).
### Custom dashboard
- Speed up generation: `smart`/**fast** instead of `smart`/slow (both
`create-dashboard-preview.tsx` and `chat-adapters.ts`).
- Pin `@babel/standalone` to `7.29.7` in the sandbox host (avoid
surprise breakage from `latest`).
- Disable analytics in generated dashboards.
### Misc
- Bump MCP RPC timeout 15s → 45s (`apps/skills/src/mcp-wrapper.ts`).
## Testing
- `pnpm typecheck` ✅
- `pnpm lint` ✅
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Refreshes the model selection matrix, forwards auth headers so template
rewrites use authenticated tiers, and speeds up custom dashboard
generation with a more stable sandbox.
- **Refactors**
- Update model IDs: `openai/gpt-5.5`, `google/gemini-3.5-flash`,
`z-ai/glm-5.2`, `nvidia/nemotron-3-super-120b-a12b`.
- Use `openai/gpt-5.5` for authenticated fast routes.
- Forward `x-stack-*` / `x-hexclave-*` headers; build via Map to avoid
prototype-pollution; inner generate call uses the authenticated tier.
- Lower email template rewrite quality to `dumb`/`slow`.
- Switch dashboard generation to `smart`/`fast` in
`create-dashboard-preview.tsx` and `chat-adapters.ts`.
- Disable analytics in generated dashboards.
- Bump MCP RPC timeout from 15s to 45s.
- **Dependencies**
- Pin `@babel/standalone` to `7.29.7` in the sandbox host.
<sup>Written for commit 94354ae0f6.
Summary will update on new commits.</sup>
<a
href="https://cubic.dev/pr/hexclave/hexclave/pull/1615?utm_source=github"
target="_blank" rel="noopener noreferrer"
data-no-image-dialog="true"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://www.cubic.dev/buttons/review-in-cubic-light.svg"><img
alt="Review in cubic"
src="https://www.cubic.dev/buttons/review-in-cubic-dark.svg"></picture></a>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
## Release Notes
* **Performance**
* Improved AI generation speed for dashboard creation and related chat
flows by using faster AI routing.
* Increased MCP JSON-RPC request timeout to better handle long-running
operations.
* **Technical**
* Template rewriting with AI now forwards authentication-related headers
to downstream AI calls for more consistent authorized behavior.
* Updated AI model routing/selection used by the proxy layer.
* **UI/Integration**
* Pinned the sandbox Babel CDN script to a specific version.
* Disabled analytics in the sandbox SDK configuration.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
