Application Platforms
Managed hosting for web apps and APIs — push code, skip the ops.
SatJun 27, 20267 releases
Daily summary
AI SDK patches memory bug, adds Fireworks tier and MCP completions
The AI SDK fixed an unbounded memory issue in JSON response body reads, added a serviceTier option for Fireworks chat models, and enabled MCP client support for server completions, while Vercel CLI now applies per-service functions config to experimental V2 services.
- AI SDK limits JSON response body reads to prevent unbounded memory use across provider-utils packages
- AI SDK adds serviceTier provider option for Fireworks chat models and allows MCP client to use server completions
- Vercel CLI now honors per-service functions configuration in vercel.json for experimental V2 services at build time
FriJun 26, 202613 releases
Daily summary
AI SDK expands web search and MCP session support
The AI SDK added web search results access across OpenAI providers, MCP gained Streamable HTTP session reattachment, and Vercel's Python runtime now supports WebSockets for WSGI apps.
- AI SDK OpenAI providers now support the web_search_call.results include option, giving access to search results alongside generated content
- AI SDK MCP added Streamable HTTP session hooks and detach-on-close support for reattaching to sessions
- AI SDK Gateway gained a has provider option to restrict routing by model capabilities and added Exa search tool support
- Vercel Python Runtime supports WebSockets for WSGI apps like Flask via flask-sock, exposing the raw connection socket for upgrade requests
- Vercel CLI migrated service auto-detection to V2 format and deprecated the synchronous getVercelOidcTokenSync method
Vercel12