Open source AI agent platforms, compared on what you own

What each platform actually is

Run this search and you meet four different products at once. n8n is workflow automation with AI nodes. Dify is a visual builder for AI apps. CrewAI is a Python framework. Kortix is an AI Operating System that runs a company's whole agent workforce. There is also e2b's awesome-ai-agents list on GitHub, a catalog of agent projects with more than 30,000 stars as of October 2026. It maps the field without comparing it.

The dimensions that decide a purchase rarely show up in roundups: where your configuration lives, whether self-hosting is actually free, whether any model can run on your own API keys, and how finished work gets approved. Those four answers split the field, and the comparison chart on this site scores more tools on them.

Kortix: the whole layer as files

Kortix is the open-source AI Operating System: your agents, their skills, your company memory and every connector in one git repo you own. Agents, skills, memory, connector wiring, triggers and the machine image are all text in that repo, versioned and diffable. Pick any model provider and use your own keys, or connect the ChatGPT plan you already pay for. Work runs on real cloud computers: one isolated sandbox per session, on its own branch, cloning the repo and committing what it did. Nothing merges itself: work reaches main through a change request a person approves. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free.

The connectors carry real work: 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP, and connector credentials are brokered server-side, never entering the machine. Governance is written where a security reviewer can read it: per-resource permissions for people and agents, roles and an audit trail, secrets encrypted at rest, and approval gates you set, off until you set them. Free costs $0 with 200 credits a month and 1 project; the team plan is $40 per seat per month with 2,500 credits per seat, pooled. Kortix has 20,000+ GitHub stars as of October 2026, and the self-hosting guide on this site walks the install.

The repo, in one glance:

acme-co/                     # your whole company, as files├── agents/                  # who does the work├── skills/                  # how your company does a job├── memory/                  # what it has learned so far└── kortix.yaml              # rules, triggers, connectors, machine image

The workflow and framework tools

Dify positions itself as the platform for production-ready agentic workflows: build agentic workflows and RAG pipelines with a visual builder, then deploy on Dify's cloud, in your VPC, or self-hosted. Configuration lives on that visual canvas, inside the Dify workspace, so the logic is readable and the platform holds it. The Community edition self-deploys with Docker, and Dify's own pricing page markets it as open-source with paid plans for enterprise features; a free Sandbox tier starts you with 200 message credits (checked October 2026). Included credits cover models from OpenAI, Anthropic, Gemini and others; when they run out you switch to your own API keys. It fits teams shipping customer-facing AI apps, chat assistants and document pipelines, where the canvas is a feature and the platform holding the configuration is acceptable.

n8n is an AI workflow automation platform for technical teams, with a standard self-hosted version free on GitHub, per its pricing page; the front page invites you to deploy with Docker and read the entire source code. Workflows live on the n8n canvas, with JavaScript or Python steps where you need them. Watch one boundary on self-hosting: Business and Enterprise features on your own instance require a paid key that must check in with n8n's servers daily to stay active, per n8n's own pricing page. Cloud plans start at $20 a month billed annually (checked October 2026), and the Starter plan includes AI models without API keys. It fits ops and IT teams automating business processes where AI is one step among many.

CrewAI is an open-source Python framework for orchestrating autonomous AI agents, organized as Flows that define the process and Crews of role-playing agents that execute tasks. Your configuration is Python code; CrewAI's visual editor exports to it. On top of the framework sits a paid enterprise runtime that deploys on CrewAI's cloud, your own VPC, or your own infrastructure, with a control plane in the execution path of every workflow: RBAC, immutable audit trails, human-in-the-loop approval gates, and multi-LLM model swapping at runtime. It fits engineering teams writing multi-agent Python programs who want governed operation on whichever deployment they choose.

Approval is where the deepest difference sits. Dify can pause a workflow to ask for human input, n8n can require approval on individual tool calls, and CrewAI gates execution with human-in-the-loop checkpoints. Each approves a step. Kortix reviews the work itself: a change request is a diff of everything the agent did, and a person merges it or sends it back.

The other names you will meet

OpenHands is an open platform for agentic software development: its pricing page describes a full, free version you run locally with your own model key, a free cloud tier, and an enterprise option that self-hosts in your VPC. It is model-agnostic and built for bringing your own keys. It fits software teams automating development work such as pull request reviews, incident triage and CI fixes; company-wide work outside the codebase stays outside its scope.

Dust is hosted software for human-agent collaboration: 70+ connectors, 20+ frontier and open-source models from OpenAI, Anthropic, Google and Mistral, permissions with audit logs, and a dedicated single-tenant deployment for companies with strict data requirements. Its site documents no self-host path; the company publishes a repository on GitHub, and the product you buy runs on Dust's infrastructure. It fits companies that want a polished managed assistant layer and accept the vendor holding the stack.

Budibase builds internal tools, automations and agents on your own data, LLMs and APIs: employees ask questions or request approvals in Slack or Discord, an agent drafts the answer or the ticket, and apps and tables track the work. Its pricing page lists a fully open-source self-hosted plan, free to use indefinitely, with paid cloud plans from $19 a month (checked October 2026). It fits teams digitizing internal operations such as approvals, support triage and small apps, with AI woven in rather than leading.