Sease at Search Solutions 2026

This tutorial explores the paradigm shift from traditional query-response information retrieval to agentic search, focusing on the open-source ecosystem in 2026. 

An AI search agent is an autonomous entity that interprets intent, decomposes complex information needs, executes multi-step retrieval actions, and synthesises answers. 

Unlike human-driven search—which relies on manual query reformulation and trial-and-error—agentic search automates these cognitive loops, interacting natively with APIs and data stores. 

Consequently, evaluation metrics must pivot from classic relevance scores like NDCG to agent-specific benchmarks, including step efficiency, tool-use accuracy, and factual alignment. 

However, autonomy introduces severe vulnerabilities, making robust security guardrails—such as prompt injection mitigation, strict data-access boundaries, and runtime execution constraints—critical for safe deployment.

In 2026, the open-source landscape provides powerful foundations for these workflows. 

This tutorial delivers an architectural deep dive into leading search technologies supporting agentic systems.

We examine Lucene-based engines, specifically demonstrating how Apache Solr natively exposes its capabilities to large language models via the Apache Solr Model Context Protocol (MCP) server. We will show how to utilise Solr MCP to dynamically map collection schemas, query templates, and hybrid search operations as tools directly into LLM orchestration frameworks. 

Additionally, we analyse other Lucene-based engines like Elasticsearch and OpenSearch to implement the same. 

We then contrast this with Vespa, demonstrating how to exploit its native tensor operations, multi-stage custom ranking, and real-time computational graphs for advanced agentic workloads. Attendees will gain hands-on patterns to implement production-ready, secure open-source agentic search platforms.

• Delivery format: In-person only

• Target audience: 

  • Technical Managers

  • Data scientists

  • Software Engineers

  • Developers

  • Machine Learning enthusiasts

• Learning outcomes:

  • You will have a complete overview of the main aspects of agentic search

  • You will learn how to set up and configure a search engine to be used by agents, using open-source technologies

  • You will learn how to validate the performance of your agentic search, following search quality evaluation principles

  • You will learn how to set up and configure your LLM agent to make the best use of your search engine

• Tutorial schedule and description: 

Part 1: Foundations & Paradigms (90 Minutes)

• 00:00 – 00:30 | Foundations of Agentic Search

  • Defining the search agent: autonomy, planning, tool usage, and memory.

  • Transitioning from static queries to multi-step reasoning loops.

  • Cognitive load shift: Human-driven search vs. agent-driven search.

• 00:30 – 01:00 | The Open-Source Agent Orchestration Ecosystem

  • State of the art: Model Context Protocol (MCP) and orchestration (LangGraph, AutoGen, LlamaIndex, Haystack…)

  • How LLM capability improvements (reasoning models) have changed IR pipelines.

• 01:00 – 01:30 | Evaluation Metrics & Benchmarking

  • Moving beyond NDCG, MRR, and Precision/Recall.

  • Agent metrics: Step efficiency, tool-use accuracy, and trajectory correctness.

  • System metrics: Latency-to-resolution, tokens-per-grounded-fact, and cost efficiency.

 01:30 – 01:45 | Morning Break (15 mins)

Part 2: The Lucene Ecosystem Deep Dive (90 Mins)

• 01:45 – 02:05 | Elasticsearch Support for Agents (20 mins)

  • Utilising Elasticsearch for managing agent conversational state.

  • Hybrid lexical-vector matching capabilities and payload scoring for tool outputs.

• 02:05 – 02:35 | OpenSearch Native Agentic Framework (30 mins)

  • Deep dive into OpenSearch’s built-in Agentic Search engine framework.

  • Registering LLMs and creating autonomous agent tools directly inside the cluster boundary.

• 02:35 – 03:15 | Apache Solr as a First-Class Agent Citizen (40 mins)

  • Deep dive into the Apache Solr Model Context Protocol (MCP) server.

  • Architectural blueprint: How Solr MCP automatically exposes collection schemas.

  • Transforming Solr query templates and streaming expressions into LLM-callable tools.

  •  

 03:15 – 04:15 | Lunch Break (60 mins)

Part 3: Hands-On Solr & Vespa Deep Dive (135 Mins)

• 04:15 – 04:45 | Hands-On: Running Agentic Search with Apache Solr (30 mins)

  • Step-by-step code walkthrough connecting an LLM agent framework to Solr via MCP.

  • Handling complex multi-turn filtering, grouping, and faceting autonomously.

• 04:45 – 05:30 | Vespa Platform Architecture for Agents (45 mins)

  • Vespa MCP

  • Vespa’s real-time computational graphs and structured tensor operations.

  • Multi-stage ranking expressions designed for autonomous reranking.

• 05:30 – 06:30 | Hands-On: Running Agentic Search with Vespa (60 mins)

  • Deploying an agent that utilises Vespa’s MCP.

  • Executing asynchronous parallel tool calls inside Vespa’s stateless container layer.

 06:30 – 06:45 | Afternoon Break (15 mins)

Part 4: Security, Guardrails & Future Frontiers (45 Mins)

• 06:45 – 07:15 | Security Boundaries & Agent Guardrails (30 mins)

  • Mitigating indirect prompt injection via untrusted retrieved documents.

  • Enforcing strict user data-access boundaries inside agent-managed queries.

  • Runtime execution constraints: Timeouts, loop-breaking, and token rate-limiting.

• 07:15 – 07:30 | Wrap-up, Q&A, and Research Frontiers (15 mins)

  • Open problems in open-source agentic IR (e.g., collaborative multi-agent search).

• Tutorial logistics/materials: slides + code snippets