Google ADK is an open-source, code-first Python toolkit for building and deploying AI agents, optimized for Gemini. Kortix is the open-source AI Operating System: your agents, skills, memory, and connectors live in one git repo you own, and each session runs on its own cloud computer.
ADK's cells state what its own repository and docs say, and the Kortix column quotes Kortix.
| Criteria | What it is | Open source | What you own | Runs the agents or defines them | Self-host | Model choice | Human review of changes | Deployment and ops |
|---|---|---|---|---|---|---|---|---|
| Kortix | The open-source AI Operating System: agents, skills, memory, connectors in one repo | Yes | Agents, skills, memory, and connectors in one git repo you own | Runs them: one isolated sandbox per session, on its own branch | Self-host is free; or Kortix Cloud, your VPC, on-prem | Any model provider with your own keys | Session work reaches main through a change request a person approves | Managed cloud or your own infrastructure; every change is reviewed |
| Google ADK | A code-first Python toolkit for building AI agents | Yes | The agent code you write and the infrastructure you run | Defines them: you write agent logic and workflows in Python | Yes, on your own infrastructure or Google Cloud | Optimized for Gemini; model-agnostic for other providers | A tool confirmation flow; your code sets the gate | You deploy it on Cloud Run or Vertex AI Agent Engine |
ADK's cells come from Google's own repository and docs, checked October 2026. Verify them before you commit.
$ kortix sessions new --prompt "Reproduce issue #142, fix it, and open a change request"$ kortix cr ls$ kortix chatA session works on its own branch and opens a change request, and you read the diff before it lands.
Google ADK, the Agent Development Kit, is an open-source, code-first Python toolkit for building, evaluating, and deploying AI agents. It applies software development principles to agents: a Workflow runtime composes deterministic execution flows, a Task API handles agent-to-agent delegation, and a tool ecosystem covers pre-built tools, OpenAPI specs, MCP tools, and custom functions. ADK is optimized for Gemini while staying model-agnostic, and it ships a development UI and an evaluation command (the ADK repository). It runs on Python 3.11 and later, and its releases land roughly every two weeks.
ADK is a toolkit. You write agent logic, tools, and orchestration in Python, then run the result on infrastructure you choose. The framework gives you the graph, the delegation, and the evaluation, and the deployment and the operations stay with your team. Kortix is the leading open-source alternative to Claude Cowork and ChatGPT Work, and it sits one layer up: an open-source AI Operating System where the agents, their skills, the company memory, and every connector live in one git repo you own (Kortix on GitHub). Each session runs on its own isolated cloud computer, so the agents work on a real machine.
With ADK, a team owns the agent code and the infrastructure it runs on. It writes the flows in Python, deploys them to Cloud Run or Vertex AI Agent Engine, and operates the runtime, the scaling, and the logs (the ADK documentation). On Kortix, the company owns the configuration in the repo, and the platform runs the sessions: one isolated sandbox per session, connector credentials brokered server-side and never entering the machine, and per-resource permissions for people and agents.
An agent toolkit answers how to build and evaluate an agent. A platform answers where the agents run, what each one may touch, who approves a change, and how the work reaches the codebase. Teams move to a platform when the toolkit stops being the hard part, and they need a preview per change, scoped reach into 3,000+ apps, and a human approval gate before merge. ADK keeps its strength for teams already on Google Cloud that want graph-based agents inside their own code. Whether the layer you need is a framework or a platform is the question behind the framework or platform page, and the Kortix vs LangGraph page compares one framework with Kortix directly.
Kortix replaces the runtime layer, not the toolkit's role in your code. ADK defines agents, workflows, and evaluation inside a Python project you deploy; Kortix runs agents on isolated cloud computers, keeps the configuration in a repo you own, and lands changes through a review. A team building graph-based agents for Gemini can keep ADK.
ADK is a code-first toolkit with a graph workflow runtime, a Task API for delegation, and built-in evaluation, and it runs inside a Python project you deploy. Kortix runs agents as finished work on managed machines rather than as a toolkit in your code. When the agent logic belongs inside the product you ship, ADK fits that shape.
Yes. Kortix runs on Kortix Cloud, in your VPC, or on your own on-prem network, and self-host is free. The first run asks six things and no model key. Read the docs walks through it.
Yes. Kortix is open source. Read the code, fork it, and audit it at Kortix on GitHub, and self-host the whole stack on your own infrastructure.
One repo for the agents, the skills, and the memory.