Changelog

✨ September’s spotlight is an agent-ready web stack. Firecrawl turns sites into clean, AI-ready data: scrape a page to markdown or HTML, extract with a prompt or schema, crawl a domain, map URLs, and flag page changes. Tavily handles the LLM half: web search with an optional generated answer, extract or crawl pages, and deep research with sourced reports. Jina adds the rest of the pipeline: read a URL into markdown, search, chunk, embed, rerank, classify, and run DeepSearch.

Also this month: Retell AI for outbound voice-agent calls, transcripts, and post-call analysis; Creatomate for template and RenderScript video, image, and GIF renders; TikTok and Bluesky for creator publishing; Baserow for no-code table CRUD and row-change webhooks; plus Supabase for Postgres rows, storage, and database webhooks, and Qdrant for vector upsert and similarity search.

🔧 Enhancements lean on schemas, auth, and media ops: document version and relationship actions in Veeva Vault; On-prem NTLM and expanded related-object field picks in OData; clearer SMTP send-failure reporting in Email (Custom) OPA; $select-scoped pills and $expand in SAP Sales Cloud; Google Vertex AI Generate content (Gemini) in place of deprecated PaLM 2; batch Create in Amplify; runtime Client secret preservation for VUA and opt-in full HTTP error bodies in OpenAPI; and Serverless as a job type in UiPath.

🛡 Stability fixes cover OpenAPI custom request headers coerced to text, UiPath Start Job targeting (Invalid RobotID) in modern folders, SAP Business One Search retries that keep the object name, AWS Secrets Manager Secret string sent as plain text, and Azure File Storage Upsert so large uploads no longer stop after the first 4 MB.