Elastic Context Engine

Context that's complete and fresh

Elastic Context Engine* connects any enterprise source to build accurate agents: sync context every turn or on a schedule, precompute for fast cross-source retrieval, and constantly improve accuracy.

Learn more
Get access

*Experimental

Read our blog to learn how to continuously curate and manage context.

Read the blog

Visit our documentation to learn more about Context Engine and how you can get started.

View docs

What does a context engine do for agents?

Precompute context to cut input tokens, reduce latency, and improve answer quality. Using Context Engine for support agents reduced cost by 55% and latency by 40%.

Baseline RAG
With Context Engine (*Source)
Accuracy
0.625
→0.917 +29%
Input tokens (Total 96 prompts)
174.8M
→42.6M −75%
F1 Score
0.561
→0.827 +26%
Accuracy
Input tokens (Total 96 prompts)
F1 Score
Baseline RAG
With Context Engine (*Source)
0.625
→0.917 +29%
174.8M
→42.6M −75%
0.561
→0.827 +26%

BENEFITS

Make any agent efficient

Continuously keep your agents' context complete, fresh, efficient, and measurable.

  • Complete context that never goes stale

    Keep context current by assembling from live sources as needed. Build full context by mapping context sources up front using a wide range of external connectors.

  • Agent accuracy that compounds

    Improve accuracy with every loop as agent traces land back in the same retrieval layer the agent already queries for context, continuously improving accuracy with each agent session.

  • Slash tokens on every turn

    Use fewer tokens per query with Knowledge Indicators that map each source's schema, field semantics, and query patterns into structured metadata ahead of time. This optimizes discovery and relevance so at query time, agents get only the sharp facts they need instead of a wall of text.

  • Works with any harness

    Ground an agent in all your organization's context. Access it inside any harness, model, framework, or protocol you already build in.

How it works

Connect enterprise data, use Elastic Workflows to distill it into structured Knowledge Indicators, and serve that context to any agent.

Connect

Connect your agents to context from Elasticsearch indices, external data sources via Kibana connectors, or ES|QL queries.

Assemble

Build automations to create Knowledge Indicators that extract data semantics, descriptions, relationships, and facts into an AI index.

Retrieve

Leverage composable ES|QL to retrieve the most relevant context at inference time.

Observe and improve

Use evals and traces to track agent responses and improve context.

Integrate

Deliver context to first- and third-party agents through plugins to integrate different harness from Claude Code, Codex, LangChain Deep Agents, Claude-managed agents, AWS AgentCore, Gemini Agent Platform, and others.

Architecture

Automations pre-extract Knowledge Indicators into an AI index, so any agent can retrieve relevant context with a simple ES|QL query.

Context types

Create several types of Knowledge Indicators from raw data.

Semantic metadata from sources

Derive profiles of data sources including descriptions of purpose, fields, data distribution and characteristics to instruct how to query and when to use.

Entities and relationships from documents

Extract entities like accounts, contacts, projects, business objectives, systems, and services, and store them along with semantic relationships to guide agent reasoning.

Document facts and summaries

Summarize large, complex documents and extract key facts to reduce expensive document reads.

Agent memory from interactions

Capture memories to fast-track responses and give every agent compounding knowledge.

USE CASES

Agent use cases

Map, precompute, and identify relationships.

  • Operations Analysis Agents

    Create the context map that allows agents to trace across multiple sources of logs and events to find gaps and failures and diagnose issues reliably.

  • Knowledge Base Agents

    Precompute facts that help agents shortcut common questions even as the data and types of questions change.

  • Investigation Agents

    Identify relationships within the data that lead to non-obvious findings and insights that can generalize solutions and find new opportunities.

INTEGRATIONS

Best in class? Built right in

Integrate with the most popular agent frameworks and applications.

RESOURCES TO START

Take the next steps

Learn more with walkthroughs, demos, and documentation.

  • Walkthrough

    Follow a step-by-step walkthrough to build and store context in an AI index and power smarter agents using fewer tokens.

    Walkthrough

  • Demo

    Explore this interactive demo to see how easy it is to get started with Context Engine.

    Interactive demo

  • Docs

    Learn more about Context Engine and how you can get started.

    Documentation

Frequently asked questions

What is a context engine?

A context engine is a part of agent architecture that gathers and builds context from business-specific data and interactions to make an agent more efficient and effective.

What's the difference between a context engine and a vector database or memory store?

A context engine builds on vector databases and memory stores. A vector database acts as a storage and retrieval component for context, while a memory store captures context directly from agent interactions. A context engine decides what to put in those stores; how to structure it; how to maintain, govern, and secure it; and the tools to expose it for retrieval.

I already have a data catalog. Why do I need a context engine?

Data catalogs describe data but they rely on static information and need to be kept up to date manually. As data changes, agent use cases expand, user behavior and prompts change, and data catalogs quickly get stale. They also typically lack insight from within the data, and they can describe data shape and schema, but not entities, anomalies, or trends that require search and retrieval.

What is a Knowledge Indicator?

A Knowledge Indicator is a unit of context. It can be any piece of context that may be useful to agents, including index and data source descriptions, extracted entities, document-based facts, and more.

What's stored in an AI index?

An AI index stores knowledge indicators derived from data sources and memories derived from agent interactions and traces.

How is security and permissions enforced?

AI indices can be managed and permissioned just like other Elasticsearch indices with full role-based access control. Workflows and agents can be run as a user to limit and audit access. Context Engine also captures the provenance of every Knowledge Indicator generated so access can be managed.

How is Context Engine priced?

Context Engine is currently not priced separately while under Private Preview. Pricing will be announced at a later date.