ReAct Engine
Thought→Action→Observation loop with 6-stage context compaction and intelligent token management.
Appearance
Production-grade Agent infrastructure. SPI-driven. Reactive inside, pragmatic outside.

Thought→Action→Observation loop with 6-stage context compaction and intelligent token management.
16 SPIs with @Stable / @Experimental annotations. 119 stable types, japicmp-enforced compatibility.
Built on Project Reactor. Non-blocking throughout, but pragmatic blocking APIs when you need them.
Circuit breaker, loop detection, cooperative cancellation, permission guards, PII redaction, and JDBC audit trail.
Claude Code format compatible. Install plugins from local, GitHub, Git, or npm. Skills, hooks, MCP, agents.
Multi-channel routing, session management, streaming, and mirroring. DingTalk shipped, Slack/Feishu planned.
Google ADK-compatible Agent-to-Agent communication. AgentCard discovery, in-process and remote invocation.
A production coding agent CLI built entirely on Kairo — 56 tools, skill system, multi-agent teams, IDE bridge.
Self-hosted multi-channel personal AI assistant — 50+ tools, 8 messaging channels, self-evolution, OpenAI-compatible API.
ExecutionSandbox SPI for safe command execution. LocalProcess default, Docker/K8s pluggable.