Kairo Code

Kairo Code is a production coding agent CLI built entirely on the Kairo framework. It serves as the primary dogfooding application — proving that Kairo's SPI-driven architecture, reactive runtime, and agent infrastructure work under real-world coding workloads.

Source code: github.com/CaptainGreenskin/kairo-code

Why Kairo Code?

Kairo Code is not a fork of any existing coding agent. It is a Java-native coding agent built from the ground up on Kairo's Agent OS abstractions:

• Framework validation — Every feature in Kairo Code exercises a Kairo SPI. If something is missing or poorly designed, we fix the framework first. This feedback loop keeps the framework honest.
• Java all the way down — The agent runtime, tool execution, context management, model routing, and session persistence are all Java, running on Project Reactor. No shelling out to Python or Node for core logic.
• Production-grade by inheritance — Kairo Code inherits Kairo's circuit breaker, loop detection, cooperative cancellation, 6-stage context compaction, and permission guards without reimplementing them.
• Extensible by design — Skills, tools, model providers, and stateless Agent Profiles are pluggable via Kairo's SPI contracts. Adding a new tool or swapping a model provider requires no changes to the agent core.

Screenshots

CLI Demo

REPL Startup

Tool Execution — Grep Search

Skill System — Load & Code Review

Plan Mode — Read-only Analysis

Cost Tracking

Web UI

Welcome & Setup

Chat Interface

Workspace — Explorer, Search, Source Control

Self-Evolution & Hook Configuration

Light Theme

Desktop App

Kairo Code also ships as a desktop workspace with an integrated chat agent, file explorer, source control, and self-evolution tooling.

Main Chat Interface

Explorer Sidebar

Source Control

Self-Evolution

Kairo Code learns from each session — accumulating memory and knowledge, and refining behavior through the hook system:

Approval Mode

Control how tools run with Manual, Auto-safe, and YOLO approval modes, backed by hook-based safety controls:

Light Theme

Agent Workspace

A full IDE-like workspace — chat agent, file explorer, source control, and theming working together:

Key Features

Feature	Description
Interactive REPL	Streaming output, tool approval prompts, rich terminal rendering
Extensible Tool Runtime	Framework and product tools for files, execution, web, Git, interaction, planning, memory, scheduling, and code intelligence
Skill System	Markdown-based extensible capabilities loaded at runtime
Agent Collaboration	Parent-owned, task-scoped Subagents with independent context, steering, cancellation, and evidence-based acceptance
IDE Bridge	WebSocket-based Bridge SPI for VS Code, JetBrains, and Zed integration
Session Persistence	FileSessionStorageProvider for conversation continuity across restarts
Multi-Model Support	Claude (native Anthropic), GLM, Qwen, GPT (OpenAI-compatible)
Plan Mode	Enter/exit plan mode for structured multi-step reasoning
Memory	Read/write/delete persistent task and project knowledge
Cron Scheduling	Create, manage, and trigger recurring tasks

Tool Categories

Kairo Code combines Kairo's focused framework tools with product-scoped tools. The registry is assembled from enabled capabilities rather than a fixed marketing count:

• File Operations — Read, Write, Edit, Glob, Grep, Tree, Diff, BatchRead, BatchWrite, SearchReplace, PatchApply, JsonQuery, TemplateRender
• Execution — Bash, Monitor, Mvn, Sleep, VerifyExecution
• Web — WebFetch, WebSearch, Http, OpenApiHttp
• Git — Git, Github
• Interaction — AskUser
• Skills — SkillList, SkillLoad, SkillManage
• Agent Operations — authoritative task delegation, send_message, task lifecycle, TODO, and workflow tools owned by the live product runtime
• Plan Mode — EnterPlanMode, ExitPlanMode, ListPlans
• Memory — MemoryRead, MemoryWrite, MemoryDelete
• Cron — CronCreate, CronDelete, CronEdit, CronList, CronPause, CronResume, CronTrigger
• Code Intelligence — Lsp

Supported Models

Provider	Models	Integration
Anthropic	Claude Opus, Sonnet, Haiku	Native AnthropicModelProvider
GLM	GLM-4 series	OpenAI-compatible adapter
Qwen	Qwen series	OpenAI-compatible adapter
OpenAI	GPT-4o, GPT-4, etc.	OpenAI-compatible adapter

Relationship to Kairo Framework

Kairo Code depends on the Kairo framework as a library consumer:

kairo-code-core
  ├── kairo-core          (ReAct engine, compaction, model providers)
  ├── kairo-tools         (built-in tool suite)
  ├── kairo-mcp           (MCP protocol integration)
  ├── kairo-skill         (skill system)
  └── kairo-plugin        (plugin system)

Kairo Code assembles Kairo's SPIs and owns the interactive Task/Subagent lifecycle. Kairo supplies the reusable Agent, risk, tool, skill, and protocol primitives; the product remains authoritative for queues, permissions, cancellation, persistence, acceptance, and UI projection.