Open source AI agent platforms on GitHub, repo by repo
The fastest way to judge an open source AI agent platform is to read its repo. Kortix is the open-source AI Operating System: your agents, their skills, your company memory, and every connector in one git repo you own, with the agents working on real cloud computers. Its source lives on GitHub, and so do the alternatives'.
Real commands, real files
The quickstart, the session loop, and one agent's definition, quoted from the repo's README.
# 1 · Install the CLIcurl -fsSL https://kortix.com/install | bash# 2 · Scaffold a project — creates kortix.yaml + your agents, skills and runtime configkortix init# 3 · Ship it — pushes your repo and brings the whole thing live in the cloudkortix shipkortix sessions new --prompt "Summarize this week's commits and open a change request"kortix cr ls # review what an agent proposes — merge to keep itkortix chat # talk to a session's agent from your terminal# kortix.yamlagents: support-triage: file: agents/support-triage.md # what it does, in plain markdown connectors: all # what it may reachWhere the Kortix source lives
Kortix ships its whole platform in one public repo: Kortix on GitHub. Open source. Read it, fork it, audit it. The web app, the API, the CLI, the desktop and mobile apps, and the sandbox image sit under apps/ as web, api, cli, desktop-electron, mobile and sandbox. The documentation is authored in the repo too, at apps/web/content/docs. One version number at the root covers all of it, so API, frontend, CLI and desktop release together. The repo counts 20,000+ GitHub stars as of October 2026.
What you get from the repo
The repo is the configuration. kortix.yaml declares the machine image, connectors and triggers, and secret names and grants, and secret values never appear in the repo. An agent is agents/<name>.md (behavior) plus an agents.<name> block in kortix.yaml (governance). A skill is markdown plus scripts that encode how the company does one job, written once and shared into every session. Memory is plain files today: the living company brain, kept in the same tree, versioned and diffed with everything else. Triggers come in exactly two types, cron and webhook.
The runtime follows from that shape. One isolated sandbox per session. Each session has its own isolated machine and branch. Session work reaches main through a change request. Merge is default-deny for agents. Connectors: 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP. Any model provider with your own keys. Or the ChatGPT plan you already pay for.
How to try it
Install the CLI with the curl one-liner. kortix init scaffolds kortix.yaml plus your agents, skills, and runtime config. kortix ship brings the whole thing live in the cloud. Work starts from the terminal too: kortix sessions new with a prompt, kortix cr ls to review what an agent proposes, and kortix chat to talk to a running session.
Prefer no install? Try Kortix: create a project and start a session in the browser, nothing to install. Read the docs for the full command surface. The repo also runs on hardware you own: kortix self-host start brings up a production-style instance from Docker images. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free. The self-hosting guide on this site walks that install step by step.
The other repos a developer will meet
Kortix is the recommendation on this site, and it is not the only open source agent code on GitHub. Three more repositories cover different jobs, each described from its own README as of October 2026.
CrewAI describes itself as an open-source Python framework with high-level abstractions and low-level APIs for building production-ready multi-agent workflows. Crews handle role-based agent collaboration; Flows handle event-driven control. You install it with uv tool install crewai on Python 3.10 to 3.13, and a new project keeps agents in agents/*.jsonc and crew settings in crew.jsonc, run with crewai run. On top of the framework, the company sells CrewAI AMP: managed deployment, observability, governance, security, and enterprise support, with the Crew Control Plane free to try. The framework sends anonymous telemetry by default; OTEL_SDK_DISABLED=true turns it off.
OpenHands heads its repo as Agent Canvas, the self-hosted developer control center for coding agents and automations. It runs OpenHands, Claude Code, Codex, Gemini, or any ACP-compatible agent across local, remote, and cloud backends. It runs locally by default, with agent backends in Docker, on VMs, or within your company infrastructure, and OpenHands Cloud or Enterprise as managed options. Setup is npm install -g @openhands/agent-canvas, then agent-canvas. Its README marks the project beta, and the model story is bring-your-own: use with any LLM.
For the widest view of the field, e2b's Awesome AI Agents is a curated catalog: open-source projects in one part, closed-source projects and companies in the other, with a filterable web version at e2b.dev/ai-agents. Keep it for widening a search once the repos in this map are not enough.
How the repos differ
CrewAI is a framework: you write the process in Python and run it wherever you run code. Agent Canvas is a control center for coding agents. Kortix is the platform layer that runs the agents themselves: agents, skills, company memory, connectors, and triggers as files in one git repo you own, thousands of agents in parallel on one config, each on its own cloud computer, with work landing through a change request a human approves. If the job is a multi-agent Python program you own end to end, CrewAI fits. If it is coding agents across your team's repositories, Agent Canvas does that job. When the platform itself should be open source and yours, run Kortix.
The wider case for running this category open source is on the Kortix blog. For the buying decision, the open source AI agent platform guide on this site compares the field before you commit.