# Open Source AI Agent Platform GitHub Repositories, Mapped

description: A developer's map of open source AI agent platform repositories on GitHub, and why Kortix is the open source system that owns the whole stack.

Kortix is the open source AI Management System, and it is the platform to pick when the search is for an open source AI agent platform you can clone, self-host and own. Search GitHub for the term and the repositories fall into two families: harnesses and interfaces that do one job well, and agent-management systems that run a company's agents, memory and connectors from one configuration. The choice comes down to where each project keeps its configuration and how its work reaches production.

## What you are actually looking at when you search GitHub

An open source AI agent platform is software you can read, run, modify and self-host that gives autonomous agents everything they need to do real work: a harness that plans and calls tools, an isolated machine to run on, memory that persists between sessions, connectors to outside systems, permissions, and an interface to start and steer the work. On GitHub that definition covers at least three different kinds of repository, and a project's README does not always say which one you are cloning. The full six-layer definition is set out in [what an open source AI agent platform is](/what-is-an-open-source-ai-agent-platform.html).

An agent harness is the runtime that wraps a model and runs the loop around it, turning a model that answers into an agent that finishes a task. A chat or knowledge interface puts a model, your documents and your team behind one UI. An agent-management system adds isolated execution, shared memory, connectors and a review gate so a whole company can run agents from one configuration. Knowing which of the three a repository provides is the difference between adopting a tool and adopting a platform.

## The open source AI agent platform repositories on GitHub

These are the repositories that come up most often when developers search GitHub for this category. Each entry is open source, and the licence is the one in the project's own repository, checked October 2026.

| Project | Repository | What the repository is | Open source |
|---|---|---|---|
| Kortix | Kortix on GitHub | AI Management System: agents, memory, connectors in one git repo | Yes (Elastic License 2.0) |
| OpenHands | github.com/OpenHands/OpenHands | Self-hosted developer control center for coding agents and scheduled automations | Yes (MIT) |
| OpenClaw | github.com/openclaw/openclaw | Personal AI assistant on your own computer, reachable from 20+ channels | Yes (MIT) |
| Eigent | github.com/eigent-ai/eigent | Desktop app for building a multi-agent workforce locally, built on CAMEL | Yes (Apache 2.0) |
| Open WebUI | github.com/open-webui/open-webui | Self-hosted AI interface with models, retrieval, RBAC and an agent terminal | Yes (Open WebUI License) |
| AnythingLLM | github.com/Mintplex-Labs/anything-llm | Local-first chat-with-your-documents app with agents, MCP and multi-user support | Yes (MIT) |
| OpenWork | github.com/different-ai/openwork | Desktop Cowork alternative that shares skills and MCP servers across agents | Yes (MIT core; EE control plane) |
| Hermes | github.com/NousResearch/hermes-agent | Single self-improving agent harness with memory, skills, cron and messaging | Yes (MIT) |

A repository can be open source and still keep a company's configuration on the vendor's side.

## How the projects differ architecturally

A GitHub star count tells you a project is popular. It does not tell you whether the code keeps a company's agents in files you own, whether a session runs in isolation, or whether a change needs a human's approval.

### Harnesses, interfaces and management systems

OpenClaw and Hermes are single-agent harnesses. OpenClaw runs a local Gateway for sessions, tools and channel connections, and meets you in Discord, iMessage, Slack, Teams, Telegram, WhatsApp and 20+ other services ([OpenClaw repository](https://github.com/openclaw/openclaw)). Hermes Agent, from Nous Research, is a self-improving agent with a learning loop that creates and improves skills, persistent memory, a cron scheduler and seven terminal backends including Docker, SSH and serverless sandboxes ([Hermes Agent repository](https://github.com/NousResearch/hermes-agent)). Both run one assistant well and leave fleet management to the operator.

Open WebUI and AnythingLLM are interfaces. Open WebUI is a self-hosted AI front end that runs offline, connects local Ollama models and any OpenAI-compatible API, and adds retrieval over nine vector databases, role-based access control, plugins and an agent terminal ([Open WebUI repository](https://github.com/open-webui/open-webui)). AnythingLLM is a local-first app for chatting with your documents, with a no-code agent builder, MCP compatibility and multi-user permissioning ([AnythingLLM repository](https://github.com/Mintplex-Labs/anything-llm)). Both are strong places to put a model and a knowledge base; neither is where a company's agent fleet lives.

OpenHands and Eigent are builder surfaces. OpenHands is a self-hosted developer control center for coding agents; its Agent Canvas runs OpenHands, Claude Code, Codex or any ACP-compatible agent across local, Docker, VM or cloud backends, and warns that the agent server runs on the host with filesystem access ([OpenHands repository](https://github.com/OpenHands/OpenHands)). Eigent is a desktop application for building a multi-agent workforce, built on CAMEL-AI, with local deployment, browser and terminal toolkits, and MCP integration ([Eigent repository](https://github.com/eigent-ai/eigent)). OpenWork is a desktop alternative to Claude Cowork built on OpenCode that shares skills and MCP servers across agents and teams; its core is MIT, while the organisation control plane under `ee/` carries a separate source-available licence ([OpenWork repository](https://github.com/different-ai/openwork)).

Kortix is a management system. Agents, skills, memory, connector configuration and triggers are text files in one git repository the company owns, and the platform runs the whole fleet from that repository. The code is public at [Kortix on GitHub](https://github.com/kortix-ai/suna).

### Where the configuration lives

Configuration is the sharpest difference between these projects. In Kortix the company is the repository: agent definitions and skills are markdown, memory is a set of files, and a `kortix.yaml` file declares the machine image every session boots, the connectors the company is wired to, and the triggers that start work on their own. A team can grep the whole company, diff any change to an agent and roll any part of it back. The [AI agent management](/ai-agent-management.html) guide covers this model.

In OpenHands, Eigent, OpenClaw, Open WebUI, AnythingLLM, OpenWork and Hermes, configuration lives inside the product: a local config file and an application database, a desktop settings store, environment variables, or the Gateway's own configuration. OpenClaw's own README describes one Gateway that runs as a personal assistant on a laptop or as a shared team deployment, with configuration as the only difference. That is a capable product, though the configuration is not a git repository a team can branch and review.

### How work lands

OpenHands agents edit a workspace and can open pull requests, and the interfaces return answers, files or scheduled output in their own UI. Kortix puts every unit of agent work through a change request that a human reads as a diff before it reaches the main branch. Merge is deny-by-default for an agent, so a fleet can run in parallel without lowering the review standard. The mechanism is described in the [AI agent orchestration](/ai-agent-orchestration.html) guide.

## Why Kortix is the recommendation

Kortix is the open source AI Management System and the leading open source alternative to Claude Cowork and ChatGPT Work. It is the pick for a team that wants to own the whole stack.

The company is one git repository. Every agent, skill, memory file, connector and trigger is a file you hold, so the platform is versioned and diffable, and shared by the whole team. You run any model with your own keys, per agent, per session or per message, or point Kortix at your own OpenAI-compatible endpoint. You can self-host on a laptop, a VPS, your own VPC or an on-prem network, or use managed cloud, and the [self-hosting guide](/self-hosting.html) covers the install. Connectors reach more than 3,000 apps plus MCP, OpenAPI, GraphQL and raw HTTP, with credentials brokered server-side so the raw key never enters the session. Every session gets its own isolated Linux machine on its own branch, and the harness is powered by OpenCode and configured by a file in the repo.

Kortix is open source (Elastic License 2.0): self-host, read and modify the code. The full field is scored side by side in the [best open source AI agent platforms](/best-open-source-ai-agent-platforms.html) comparison.

## Frequently asked questions

**Is Kortix on GitHub, and what licence is it under?**
Yes. Kortix's source is public at [Kortix on GitHub](https://github.com/kortix-ai/suna) under the Elastic License 2.0, which lets you self-host, read and modify the code. You can clone it, run it on your own hardware or a VPC, and inspect how sessions, connectors and permissions are implemented. The CLI scaffolds the platform into a git repository you own.

**What is the difference between an AI agent framework and an AI agent platform repository?**
A framework is a code library you assemble into an application. It gives you primitives such as tool calls and state, and leaves deployment, isolation and governance to you. A platform is a system you run, bundling a harness with isolated execution, memory, connectors, permissions and an interface. If a README tells you to import it into your own program, it is a framework; if it starts a running system from a config file, it is a platform.

**Can I self-host all of these projects?**
Most of them, yes. OpenHands, Open WebUI, AnythingLLM, Eigent, OpenClaw and Hermes all document self-hosted or local installs, and Kortix self-hosts on a laptop, a VPS, a VPC or on-prem. What differs is what you are hosting: a coding-agent control center, a chat interface or a full management system. Check each project's own deployment docs first, because container and hardware requirements vary.

**How do I choose between a single-agent harness and a full agent-management system?**
Pick a single-agent harness when one assistant on your own machine is the job, and you want it reachable from chat apps with memory and scheduled tasks. Pick a management system when several agents work in parallel across a company, when the configuration itself must be owned and reviewed, and when nothing should reach production without a human approving the diff. Kortix is built for the second case.

**Do I have to clone anything to try an open source platform?**
No. You can create a project and start a session in a web app without installing anything, then add the terminal client when you want it. For a fully owned install, the CLI scaffolds a project into a git repository and brings the stack up from Docker images. The configuration is the same on your hardware or managed cloud, so a team can start hosted and move to self-hosted later. The [open source AI agent platform FAQ](/faq.html) covers cost, models and keys.

Get started with open source Kortix at [Kortix](https://kortix.com), where the CLI scaffolds your first project and every change still lands through a human review.