About dkod
The dkod engine is open source (MIT licensed) — a platform purpose-built for AI agents to write, review, and ship production code safely, fast, and at scale.
Software development is being reshaped by AI agents that can reason about code, implement features, and fix bugs autonomously. But today's developer tooling was designed for humans — git workflows, CI pipelines, code review UIs — none of it was built with agents in mind.
dkod is the infrastructure layer that bridges that gap: an agent-native protocol and platform where AI agents are first-class participants, not awkward guests.
Meet the founder
Haim Ari
AI Engineering & DevOps Lead Architect at Start.io
Founder of dkod. 15+ years as a DevOps engineer and team lead in ad tech, designing and operating large-scale production systems that handle millions of requests per second with low latency. Led infrastructure teams across Kubernetes and AWS — building CI/CD pipelines, multi-datacenter deployments, and auto-scaling architectures that keep real-time bidding platforms running at scale.
Now focused on AI engineering and the agentic development ecosystem. Built dkod after seeing firsthand what happens when you point multiple AI agents at the same codebase — the tooling wasn't there. The same instincts that drove building reliable infrastructure at millions RPS now drive building reliable infrastructure for AI agents writing code.
Writes about agentic development, Claude Code workflows, and AI engineering methodology at Vantage Academy.
Why agent-native?
Retrofitting AI agents onto human-first tools creates friction at every layer. dkod removes that friction by designing for agents from the ground up.
Human-first tools
- Agents simulate keystrokes and clicks
- CI pipelines take minutes per run
- Code review requires human context switches
- Merge conflicts handled manually
Agent-native platform
- Agents connect via structured protocol
- Verification completes in seconds
- AI review is built into the pipeline
- Semantic merging resolves conflicts automatically
What we build
A layered architecture where each component is designed for agent-speed operation and developer-grade reliability.
Agent Protocol
A unified interface for any AI agent to connect, authenticate, and operate on codebases.
Semantic Engine
Code intelligence layer with AST parsing, call-graph analysis, and semantic search.
Verification Pipeline
Automated lint, test, type-check, and AI review — every change verified before merge.
Session Isolation
Each agent operates in an isolated workspace overlay. No conflicts, no corruption, no surprises.
Our principles
Transparent by Default
Every agent action is logged, reviewable, and auditable. No black boxes — every code change has a clear provenance trail.
Safety as Architecture
Session isolation, semantic merging, and verification pipelines are built into the protocol — not bolted on as an afterthought.
Speed Without Compromise
Sub-50ms agent handshakes. Verification in under 30 seconds. Agents should never wait for infrastructure.
Open Core, Open Protocol
The engine and protocol are MIT-licensed. Build on top of it, fork it, self-host it — no lock-in, no vendor traps.
Join the community
Connect with developers building with AI agents at scale. Get help, share what you've built, and shape the future of agent-native development.
Discord
Chat with the team and community. Get help with setup, share feedback, and discuss multi-agent workflows.
Twitter / X
Follow for updates, architecture deep dives, and behind-the-scenes on building the agent-native code platform.
GitHub
Star the repo, open issues, contribute code. The dkod engine and plugin are open source under the MIT license.
FAQ
An agent-native code platform is infrastructure purpose-built for AI agents to work with source code. Instead of retrofitting human developer tools (git CLI, web UIs, CI pipelines) for agent use, an agent-native platform provides a structured protocol, isolated sessions, semantic code understanding, and automated verification designed specifically for how agents operate.
dkod uses session isolation and semantic merging. Each agent operates in its own isolated workspace overlay, so agents never interfere with each other's work. When agents submit changes, dkod's semantic engine understands code structure at the AST level and can automatically merge non-conflicting changes — even if they touch the same file. True semantic conflicts are flagged for resolution.
Yes. The dkod engine is MIT-licensed and designed for self-hosting. You can run it on your own infrastructure with full control over your data and code. The Pro and Enterprise plans offer managed hosting for teams that prefer not to operate the infrastructure themselves.
Built in the open
The dkod engine is MIT-licensed. Explore the source, open an issue, or start contributing.