Introduction

dkod is the first source code platform built for AI agents. It replaces the Git-based workflow with a semantic, agent-native protocol where AI agents connect, write code, get verified, and merge — in seconds, not hours.

The Problem: Git Was Built for Humans

Every tool in today's development stack — Git, GitHub, CI/CD, code review — was designed for a world where humans read files, write diffs, resolve merge conflicts, and review line-by-line changes.

That world is ending. In the next 3–5 years, the dominant workflow will be: humans describe intent, agents write code, the platform verifies correctness, code ships.

Today's AI coding agents (Claude Code, Cursor, Devin, Copilot Workspace) are forced to work through a human-designed protocol:

• Clone an entire repository when they need three functions
• Edit text files line by line when they're rewriting entire modules
• Generate line-based diffs when the change is semantic
• Create pull requests and wait for a human to click a green button
• Run CI pipelines designed for human-paced development

This is like forcing a self-driving car to use a steering wheel. dkod gives agents their own native interface — one that is 10–100x more efficient for machine-driven development.

What is dkod?

dkod consists of two parts:

dkod engine — the open-source core (MIT licensed). Includes the Agent Protocol, SDK, CLI, runner, and reference server. Anyone can clone, build, and run a fully functional local dkod server.

dkod platform — the managed service at dkod.io. Adds AI code intelligence (semantic search, AI code review, codebase Q&A), hosted infrastructure, and enterprise features.

Works With Your Existing GitHub Repos

You don't need to migrate anything. dkod is fully Git-compatible — it works on top of your existing GitHub repositories, your existing branches, your existing history. Nothing changes about your current setup.

Point dkod at a repo you already have. Your agents connect through the Agent Protocol, write code, get verified, and merge — all while your repository stays exactly where it is, on GitHub, with the same remotes, the same CI, the same collaborators.

This means you can test-drive dkod without any commitment. No new hosting. No data migration. No switching remotes. Just connect an agent to a repo you already work on and see what happens.

The difference is what happens after you try it. When your agents can connect in 50ms instead of cloning for minutes, when 20 agents work on the same branch without conflicts, when code goes from written to verified to merged in under 60 seconds — going back to raw Git feels like going back to dial-up.

dkod doesn't replace Git. It makes Git invisible — handling the plumbing so your agents (and you) can focus on what matters: shipping code.

The Key Insight: Two Protocols, One Codebase

dkod operates on a dual-layer architecture — one layer for agents, one layer for humans. Both interact with the same codebase, but through fundamentally different interfaces optimized for their nature.

Agents use the Agent Protocol — semantic, streaming, intent-based. Humans use a Git-compatible CLI and a web UI. Both see the same code, the same history, the same truth.

Why Agent-Native?

Traditional platforms treat agents as tools that run inside a human workflow. dkod treats agents as primary users of the platform.

• Sub-50ms connection — agents connect instantly via the Agent Protocol, no cloning required
• Semantic context — agents request exactly the code they need ("give me every function that touches the payment flow"), not entire file trees
• AST-level changesets — agents submit structural changes, not line-based diffs
• 20 agents, one branch, zero conflicts — session isolation lets dozens of agents work simultaneously on the same codebase without worktrees, without cloning the repo multiple times, without merge conflicts
• Automated verification — tests, linting, and AI review run as part of the protocol, not as a separate CI step
• 60-second end-to-end — from submit to merged, a typical changeset takes under 60 seconds vs. hours or days with the PR cycle

Component Map

Next Steps

• Quickstart — get up and running in 5 minutes
• Agent Protocol — understand the six core operations
• Session Isolation — how 20 agents work simultaneously without conflicts
• Semantic Merging — why AST-level merging eliminates false conflicts