Open-source AI agent platform or framework: which does your team need
The platform-or-framework question decides who runs your agents and what you own afterward. Kortix is the open-source AI Operating System: agents, skills, and memory in a git repo you own, each session on its own cloud computer. CrewAI and LangGraph are frameworks you build on; Dify is a platform you configure in a workspace.
Platform or framework, side by side
The two categories on the criteria a buyer checks, with each rival's facts drawn from its own repository.
- 01
Category
- 02
Open source
- 03
Where configuration lives
- 04
Who runs the agents
- 05
How changes are reviewed
- 06
What you deploy
- 07
Memory and state
Rival facts come from each project's own repository or documentation, checked October 2026. Kortix facts come from its documentation and pricing page.
What a platform reads from one file
kortix_version: 2default_agent: kortix agents: kortix: file: agents/kortix.md connectors: all secrets: all skills: all kortix_permissions: allThe manifest at the root of a project repo declares the agents and what each one may reach; those grants are deny-by-default.
A framework is code you build with; a platform is a place agents run
A framework is a library your engineers import. CrewAI calls itself an open-source Python framework for multi-agent workflows, and LangGraph calls itself a low-level orchestration framework for stateful agents. With either one, your team writes the application, chooses the host, wires the memory, and runs the process. The framework supplies the primitives, and the operational work stays with you (CrewAI, LangGraph).
A platform runs the agents for you. Kortix is the platform case: a project is a git repo you own, each session runs on its own cloud computer, and an agent's work reaches the default branch through a change request a person approves (Kortix). Dify is the other platform here, an LLM app development platform where teams build and run agentic workflows on one collaborative workspace (Dify).
What each choice asks of your team
A framework asks for engineers and infrastructure. You host the runtime, schedule the work, store the memory, and build the review path out of your existing tooling. That shape fits a product team building agents into its own software, where the agent is a feature and the code is the product. A platform asks for a configuration and a policy. You describe the agents, connect the tools, set what needs approval, and review the work that comes back. That shape fits a company that wants agents doing work across teams without assembling the plumbing first.
Where the line blurs
The categories overlap at the edges, and that overlap is where most of the confusion starts. CrewAI ships a framework and a commercial control plane around it, so part of that stack is a managed service (CrewAI). LangGraph pairs with a deployment platform for long-running workflows, so its hosting can be bought separately (LangGraph). OpenHands is a self-hosted control center for coding agents that runs the open-source OpenHands agent or third-party agents through the Agent-Client Protocol (OpenHands). The category label matters less than one question: when the agent finishes, who ran it and who owns the result.
How to choose
Pick a framework when you are building a product that contains agents and your engineers want control of every layer. CrewAI and LangGraph fit that shape. Pick a platform when you want agents to do work for the company and would rather configure them than build the runtime. Kortix fits the second shape and keeps ownership in view: the agents, their skills, and the memory are files in a repository you hold, so the setup is versioned and portable. Self-host is free, and the managed cloud prices seats and compute (Kortix).
What you keep after the decision
The durable answer is what leaves with you. A framework leaves you the code you wrote, which is the whole application. A platform leaves you the configuration and the memory, if it stores them where you can take them. Kortix stores them as files in a git repo, so the configuration and the memory travel together (Kortix). That is why a team can pick a platform over a framework even when both are open source: the platform can still be the copy the company owns.
The what an open-source AI agent platform is page defines the category, the comparison chart compares the main platforms across the same facts, and Kortix writes direct notes on both frameworks: Kortix and Kortix.
Platform or framework, in practice
Is CrewAI a platform or a framework?
CrewAI is a framework. Its own repository calls it an open-source Python framework with high-level abstractions and low-level APIs, and it ships a separate commercial control plane around that framework for managed deployment and governance.
Is LangGraph a framework?
Yes. LangGraph's own repository calls it a low-level orchestration framework for building, managing, and deploying long-running, stateful agents.
Can a framework become a platform?
Partly. A framework plus a managed control plane covers deployment and governance, as CrewAI's commercial suite and LangGraph's deployment platform show. The team still writes the application and chooses the memory model, which is the part a platform takes over.
Which one should a non-engineering team pick?
A platform. A framework asks the team to host and run the agents, while a platform runs them and gives people a place to direct the work. Kortix reaches people in a web app, in Slack, and in Microsoft Teams.
Run your agents from one repo you own
Kortix is open source. Build the company like a codebase, then bring it live.