Loading workspace insights... Statistics interval
7 days30 daysLatest CI Pipeline Executions
0bcf706b docs: refresh AI SDK comparison against ai@6.0.200
Corrections (verified against ai-sdk.dev docs, npm, and vercel/ai as of
June 2026):
- ai@6 is dual CJS+ESM, not ESM-only; note v7 pre-release exists
- ~38 first-party provider packages, not 50+
- Solid is community-maintained on the AI SDK side (frozen at an SDK-4-era
version) - now a TanStack differentiator, not a wash
- Lazy tool discovery is no longer unmatched: AI SDK 6 has Anthropic-only
tool search with deferLoading; ours remains cross-provider
- Narrow the middleware claim: experimental_transform, onToolCallStart/
Finish, prepareStep, and repairToolCall exist - the gap is the unified
app-level middleware system
- extendAdapter: AI SDK has stable customProvider()/createProviderRegistry();
the edge is literal-type narrowing vs string model ids
- Code execution: provider-hosted tools exist (Anthropic, xAI, OpenAI);
the edge is SDK-managed sandboxes you control
- addToolResult renamed to addToolOutput in v6
New advantages (each adversarially verified to have no AI SDK equivalent):
- Native AG-UI protocol section + wire-protocol table row
- Hooks for Every Activity section + generation-hooks table row
- Multi-Turn Structured Output section
- Debug Logging section + table row
- Code Mode Skills table row + LLM-writable skills prose
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>