Safety Controls in Agent Studio introduce guardrails for deploying agents in production. Configure global and per-IP rate limits to handle spikes, set limits on tokens, steps, and conversation depth to manage cost, and restrict access using domain allowlists. This helps prevent abuse, reduce unexpected spend, and ensure agents are only accessible from approved environments.

The Algolia Productivity MCP now supports Claude Code, Cursor, VS Code, Gemini CLI, and Codex. Developers can connect to Algolia (via Dashboard OAuth) directly from their preferred Al coding environment and query Algolia without switching context. Each AI assistant is recognised by the MCP and matched against a redirect URI, without needing to enter a Client ID. Click here to learn more.

Agent Studio now includes a dedicated search settings panel, letting teams configure search parameters specifically for their agents - separate from existing index settings. Defaults (hits per page, controlled attributes, facets) are set based on evaluation findings and applied automatically when connecting a new index. All settings are fully customizable.

Agent Studio's search tool has been upgraded with improved relevance, faster response times, and reduced cost per query. Agents now benefit from more accurate, context-aware results across a wider range of query types. AI-generated tool descriptions also help agents select and invoke the right search tool at the right time, improving overall agent reasoning quality.

A new Guardrails configuration in the Safety Controls tab allows defining custom categories to filter harmful or off-topic content across input, output, or both. Each category includes a description, examples, and an optional fallback response, helping teams control how agents handle sensitive or irrelevant queries. This improves response quality, reduces risk, and ensures interactions stay aligned with the intended use case. Click here to learn more.

The Agent Studio widget search bar now features an AI mode button and context-aware prompt suggestions. As shoppers type, relevant suggested prompts appear that transition them directly into the chat experience with a single click. The AI mode button provides a persistent, prominent entry point from search to natural language conversation, reducing friction for users moving between search and chat.

Agent Studio now includes a dedicated Getting Started page and Quick Setup flow for agents. A guided configuration wizard covers data source selection, prompt configuration, entry points, prompt suggestions, conversation features, and safety controls in a single flow. Pre-configured agent templates are available to reduce setup time and accelerate time to first deployment.

NeuralSearch users can now adjust the similarity score cutoff for NeuralSearch results based on the strength of the top match. Rather than applying a fixed threshold across all queries, the filter scales dynamically - returning broader results when top matches are highly relevant and narrowing results when match quality is lower. This improves result precision, reduces irrelevant results surfacing to end users, and eliminates the need for manual threshold configuration per index. Click here to learn more.

Optional filters now work with virtual replicas, enabling boosting or burying of specific results after sorting. Previously limited to standard replicas, this capability extends advanced relevance tuning to virtual replica configurations without requiring additional replica indices. Search teams can now apply promotion and demotion strategies across all sorting configurations, combining the efficiency of virtual replicas with fine-grained relevance control. Click here to learn more.

Agent Studio introduces dynamic search parameter passing for more flexible retrieval, along with new analytics for conversation depth and user feedback. It adds DeepSeek support, thread depth controls via API, new templates, and expanded feedback APIs and UI components. DocSearch now supports Agent Studio capabilities, helping improve customization, visibility, and control over AI experiences.