# OpenHands vs SWE-agent: two open-source scaffolds, and what comes after

> OpenHands and SWE-agent are two of the most-cited open-source coding-agent scaffolds, and each answers a different question. For a team deciding between them, the recommendation is Kortix, the open-source AI Operating System, for the layer that comes after a single scaffold.

URL: https://openhands-alternative.com/openhands-vs-swe-agent

## OpenHands vs SWE-agent, with Kortix beside them

Kortix is our recommendation; the OpenHands and SWE-agent columns state what each project documents about itself.

| | Kortix (recommended) | OpenHands | SWE-agent |
| --- | --- | --- | --- |
| Open source | Yes. Open source. Read it, fork it, audit it. | Yes | Yes |
| Self-host / where it runs | Kortix Cloud, your VPC, or on-prem. Self-host is free. | On your machine, in Docker, or on your own infrastructure | Locally, or as a browser demo; built for research |
| Model choice | Any model provider with your own keys. | Any LLM; bring your own | Your model of choice, set in one yaml file |
| Toolset | 3,000+ apps, MCP, OpenAPI, GraphQL and raw HTTP. | A software-agent SDK; runs Codex, Gemini, OpenCode and more | Tools to fix GitHub issues, find vulnerabilities, or custom tasks |
| Interface | Web, CLI, Slack, Microsoft Teams, mobile and API | A browser client (Agent Canvas) plus a CLI | A text-first agent-computer interface |
| Autonomy | Session work reaches main through a change request. Merge is default-deny for agents. | Automations on a schedule or a signed webhook | Leaves maximal agency to the model on one issue |
| Best for | A team that needs agents for more than code | A coding-agent hub for an engineering team | Research and a hackable reference; mini-swe-agent supersedes it |

The OpenHands and SWE-agent columns quote their own repositories and docs, checked October 2026. Verify each project before you choose.

## What OpenHands is
OpenHands is an open-source coding-agent project whose agents act on a codebase and apply changes. Its own repository describes it as a self-hosted hub for coding agents and automations: it runs OpenHands or third-party agents such as Codex, Gemini and OpenCode, and it can connect to local, remote and cloud backends ([github.com/OpenHands/OpenHands](https://github.com/OpenHands/OpenHands), checked October 2026). Its runtime executes an agent's actions inside a Docker sandbox with a bash shell and a code editor, so the agent can run code without reaching the host machine ([docs.openhands.dev](https://docs.openhands.dev/overview/introduction)). The project is model-agnostic: its docs say you can use it with any LLM, including open-weight models, and its own site reports 90.9K GitHub stars and 580 contributors ([openhands.dev](https://www.openhands.dev/), checked October 2026).

## What SWE-agent is
SWE-agent is an open-source research project from Princeton that gives a language model a text-first agent-computer interface for fixing issues in real GitHub repositories. Its paper is titled "Agent-Computer Interfaces Enable Automated Software Engineering" (2024), and the repository says it is built and maintained by researchers from Princeton University and Stanford University ([github.com/SWE-agent/SWE-agent](https://github.com/SWE-agent/SWE-agent), checked October 2026). The design is deliberately minimal: a single yaml file governs a run, and the repository describes the project as made for research, simple and hackable by design. SWE-agent's own repository now says most of its development has moved to mini-swe-agent, a descendant that has superseded it, matches its performance and is simpler; the maintainers recommend mini-swe-agent going forward. The repository reports that mini-swe-agent reaches 65% on SWE-bench Verified in about 100 lines of Python.

## Where each one fits
SWE-agent fits research and study: you want a minimal, readable agent to experiment with, or you are benchmarking models on issues from real repositories. OpenHands fits an engineering team that wants a self-hosted hub to run coding agents and automations on its own infrastructure. Both are single scaffolds: a tool for the job of writing code. A team that also needs shared memory, connectors into the tools it uses, and a schedule that runs while nobody is present is looking for a different layer, and the open-source one is Kortix. Teams that hit OpenHands' own limits can read the [OpenHands limitations](/openhands-limitations) page, and the head-to-head with Kortix is on the [OpenHands vs Kortix](/openhands-vs-kortix) page.

## What comes after a single scaffold
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. In Kortix, a project is a git repo with `kortix.yaml` at its root, and that repo holds the agents, skills, memory, connectors and triggers. A session runs one agent in its own sandbox on a new branch. When the agent finishes, it opens a change request against the project's default branch, and a person merges it ([kortix.com/docs](https://kortix.com/docs)). The stack section on this page shows that loop in three parts.

## The whole loop in three parts

A repo you own, a session on its own cloud computer, and a change request a person merges.

### 1. One git repo

Your agents, their skills, your company memory and every connector in one git repo you own.

`kortix.yaml`

```yaml
# kortix.yaml
agents:
  support-triage:
    file: agents/support-triage.md   # what it does, in plain markdown
    connectors: all                  # what it may reach
```

Source: https://kortix.com/docs/project/manifest

### 2. One agent per session

One isolated sandbox per session. Each session has its own isolated machine and branch.

`terminal`

```sh
$ kortix sessions new --prompt "Reproduce the checkout error, fix it, and open a change request"
```

Source: https://github.com/kortix-ai/suna (README.md)

### 3. A change request

Session work reaches main through a change request. Merge is default-deny for agents.

`change-request`

```
open ──(merge)──▶ merged      (terminal)
open ──(close)──▶ closed ──(reopen)──▶ open
```

Source: https://kortix.com/docs/work/change-requests

## Questions people ask before they pick one

### Is SWE-agent still maintained?

SWE-agent's own repository says most of its development effort has moved to mini-swe-agent, which has superseded it, matches its performance, and is simpler; the maintainers recommend mini-swe-agent going forward ([github.com/SWE-agent/SWE-agent](https://github.com/SWE-agent/SWE-agent)). SWE-agent still works and stays the reference implementation for the agent-computer interface, but new work happens on the smaller descendant.

### Can I run OpenHands and SWE-agent together?

Not as one workflow. OpenHands runs agents in its own Docker sandbox and can host third-party agents, while SWE-agent is a standalone research scaffold whose maintainers now point to mini-swe-agent. If you want one place to run coding work beside the rest of the business, Kortix runs each session on its own cloud computer and lands the result as a change request a person merges.

### Do OpenHands and SWE-agent replace a coding assistant?

They are different tools. A coding assistant suggests code as you type, while OpenHands and SWE-agent are agent scaffolds that take an issue and work on code across a repository. Kortix is a third layer: it runs agents that finish work, use your connected tools, and open a change request a person reviews before anything reaches main.

### Are OpenHands and SWE-agent open source?

Yes. Both are open-source projects with public code. Kortix, the option this site recommends, is open source too: read it, fork it, audit it, and self-host it for free.

## Run the whole system from one repo you own.

Start with one job and grow from there.

[Try Kortix](https://kortix.com)

Open source · Any model, your keys · Self-host, VPC, or on-prem