Koog is unclaimed —
K
Koog
A Kotlin-based framework designed to build and run AI agents entirely in idiomatic Kotlin.·docs.koog.ai
Koog is an open-source JetBrains framework for building AI agents, designed specifically for the JVM ecosystem. It offers an idiomatic, type-safe Kotlin DSL and a fluent Java builder API, and lets Kotlin developers deploy agents across JVM, JS, WasmJS, Android, and iOS via Kotlin Multiplatform. It supports multiple LLM providers, tool creation, retrieval-augmented generation (RAG), tracing/observability, and integrations with frameworks like Spring Boot and Ktor.
What it's for
Building basic AI agents with a predefined strategyBuilding functional agents defined as a lambda/loopBuilding graph-based agents with custom workflowsBuilding planner agents that iteratively build and execute a planCreating and running structured promptsRunning prompts across multiple LLM providersBuilding and registering custom tools for LLMsUsing Model Context Protocol (MCP) tools in agentsRetaining and retrieving knowledge via vector embeddings and RAGTracing and monitoring agent executionCompressing conversation history to optimize token usagePersisting and restoring agent stateStreaming LLM responses with parallel tool callsConnecting agents/services via the A2A Protocol
Features 18
- A2A Protocol and Agent Client Protocol support
- Agent persistence to restore state at specific execution points
- Built-in retries and fault tolerance
- Comprehensive, configurable tracing
- Content moderation and LLM parameter/model capability configuration
- Custom tool creation for external systems and APIs
- Graph-based workflow design for complex agent behaviors
- Idiomatic Kotlin DSL and fluent Java builder API
- Integration with Spring Boot and Ktor
- Intelligent history compression for long-running conversations
- Knowledge retrieval and memory via vector embeddings and RAG
- Kotlin Multiplatform deployment across JVM, JS, WasmJS, Android, and iOS
- LLM switching with seamless history adaptation across providers
- Model Context Protocol (MCP) integration for tool use
- Modular, composable feature system
- OpenTelemetry-based observability with Langfuse and Weave exporters
- Streaming API with real-time responses and parallel tool calls
- Structured output generation
At a glance
open sourceAPI
TypeLibrary / SDK
PlatformsAndroid, iOS
ForKotlin developers, Java developers
CompanyJetBrains s.r.o.
Integrations
Spring BootSpring AIKtorModel Context ProtocolOpenTelemetryDatadogLangfuseWeaveA2A ProtocolAgent Client ProtocolAmazon BedrockAnthropic
Resources