> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trykode.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Kode: The Verification-First AI Coding Agent

> Kode is the AI coding agent that never writes code without verifying it first. Every patch passes a 6-gate pipeline before touching your filesystem.

Kode is a verification-first AI coding agent. Unlike other AI code tools that generate and hope for the best, Kode runs every LLM-generated patch through a deterministic Go verification pipeline — catching syntax errors, hallucinated imports, broken function calls, and security vulnerabilities in under 50ms — before anything reaches your disk.

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Install Kode and run your first task in under five minutes.
  </Card>

  <Card title="Installation" icon="download" href="/installation">
    Install via shell script, NPM, or build from source.
  </Card>

  <Card title="CLI Reference" icon="terminal" href="/cli/overview">
    Full reference for every Kode command and flag.
  </Card>

  <Card title="Configuration" icon="gear" href="/configuration/kode-json">
    Customize models, budgets, blast radius limits, and more.
  </Card>
</CardGroup>

## How Kode works

Kode enforces a strict **Plan → Generate → Verify → Apply → Test** workflow. You describe a task; Kode builds a surgical context graph from your codebase, generates code, verifies it through six gates, applies only what passes, and runs your tests. If tests fail, it rolls back automatically.

<Steps>
  <Step title="Install Kode">
    Install Kode with a single command on macOS, Linux, or Windows. See the [installation guide](/installation).
  </Step>

  <Step title="Initialize your project">
    Run `kode init` in your project root to scaffold a `.kode/kode.json` config file with sensible defaults.
  </Step>

  <Step title="Run your first task">
    Use `kode run "your task description"` to generate, verify, and apply code in one step.
  </Step>

  <Step title="Explore advanced modes">
    Use [Ghost Branches](/concepts/ghost-branches) for parallel strategies, [TDD Mode](/concepts/tdd-mode) for test-first discipline, or [Daemon Mode](/guides/daemon-mode) for continuous background analysis.
  </Step>
</Steps>

## Key features

<CardGroup cols={2}>
  <Card title="6-Gate Verification" icon="shield-check" href="/concepts/verification-pipeline">
    Syntax, imports, function calls, blast radius, architecture, and security — all checked before writes.
  </Card>

  <Card title="Ghost Branches" icon="code-branch" href="/concepts/ghost-branches">
    Spawn parallel git worktrees with competing strategies. The winner is merged automatically.
  </Card>

  <Card title="Blindfold Mode" icon="eye-slash" href="/concepts/blindfold-mode">
    SHA-256 obfuscate identifiers before sending to any LLM for maximum code privacy.
  </Card>

  <Card title="TDD Mode" icon="vial" href="/concepts/tdd-mode">
    Block production file writes until a corresponding test file exists.
  </Card>

  <Card title="Loop Mode" icon="rotate" href="/guides/loop-mode">
    Full end-to-end pipeline with automatic rollback on test failure.
  </Card>

  <Card title="MCP Integration" icon="plug" href="/guides/mcp-integration">
    Expose Kode's verification engine to Claude Desktop and other MCP-compatible agents.
  </Card>
</CardGroup>