> ## Documentation Index
> Fetch the complete documentation index at: https://agentset-feat-agent-first-api-mcp.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

> Product updates and announcements

<Update label="July 2, 2026">
  ## Agentic playground chat

  The [playground](/search-and-retrieval/playground) and [hosted](/production/hosting-ui) chats now use agentic search. Instead of retrieving once before answering, the model searches your namespace in a tool-calling loop until it has enough context. The playground shows the full step-by-step trace for debugging, while the hosted interface shows end users a compact progress indicator.

  ### New

  * **Agentic search**: The model drives retrieval through `search` and `expand` tools, streaming step-by-step progress
  * **Chunk expansion**: The model can fetch the chunks around a search result when a chunk is cut off (Turbopuffer-backed namespaces)
  * **Accurate and Fast modes**: Replace the previous chat modes. Accurate (default) reranks each semantic search; Fast skips reranking and returns results directly
  * **GPT-5.5**: Added to the model picker as the new default model
  * **Cohere Rerank v4.0 Pro**: Now the default reranker
  * **Inline citations**: Source pills in answers open the retrieved chunk's text and metadata

  ### Removed

  * **Deep Research mode**: Agentic search handles multi-step questions directly
</Update>

<Update label="December 3, 2025" description="v1.0">
  ## General Availability

  Agentset is now generally available after 8 months of beta testing. Thanks to all our early users for their feedback.

  ### New

  * **Document processing**: All-new ingestion pipeline with image extraction, table detection, inline math recognition, and layout analysis. Choose from `fast`, `balanced`, or `accurate` modes
  * **Chunk viewer**: View document chunks directly in the dashboard
  * **Document download**: Download managed files from the dashboard
  * **Web crawling**: Crawl and ingest websites with configurable depth, path filters, and CSS selectors
  * **YouTube ingestion**: Ingest video transcripts with configurable language selection and metadata options

  ### Improved

  * Combined upload and URL forms into a single unified ingestion interface
  * Redesigned documents and jobs tables with creation timestamps and file size details
  * Turbopuffer is now the default vector store for new namespaces
  * Revamped model selector in the playground

  ### Fixed

  * Playground chat no longer shows empty settings
  * Email validation now works correctly when inviting organization members
  * Playground chat mode defaults to the correct setting
</Update>