open source alternatives to Devin

Summary

• Devin is operated by Cognition, and its documentation describes it as an autonomous AI software engineer that can write, run, and test code.
• Open-source options fall into two groups: coding agents you run (OpenHands, SWE-agent, OpenCode) and frameworks for building your own agents (CrewAI, LangGraph, MetaGPT). Each repository lists an MIT license.
• SWE-agent's README says development effort has moved to mini-SWE-agent, which has superseded SWE-agent.
• Databricks documents connecting coding agents, including the open-source OpenCode, to models governed by Unity Gateway, and hosting custom agents on Agent Bricks.
• We believe the right choice depends on whether you want a ready coding agent or a framework to build with, so compare candidates on your own repositories.

open source alternatives to Devin

Open-source alternatives to Devin include coding agents such as OpenHands, SWE-agent, and OpenCode, and frameworks such as CrewAI, LangGraph, and MetaGPT for building your own agents. Each repository lists an MIT license, so you can read, run, and modify the code. Devin's documentation describes it as "an autonomous AI software engineer that can write, run and test code." Evaluate any alternative on your own repositories and tasks.

What is Devin?

Devin is operated by Cognition. Its documentation says you can ask Devin to tackle tickets from your issue tracker, implement new features, reproduce and fix bugs, and build internal tools. This page describes Devin only through that documentation.

Which open-source projects are alternatives to Devin?

Project	Type	What its repository says	License
OpenHands	Coding agent and control center	Agent Canvas is a self-hosted control center for coding agents; it runs the open-source OpenHands agent and can use third-party agents	MIT
SWE-agent	Coding agent	Takes a GitHub issue and tries to automatically fix it, using your language model of choice	MIT
OpenCode	Coding agent	"The open source coding agent."	MIT
CrewAI	Multi-agent framework	Open-source Python framework for multi-agent workflows, with Crews for autonomous agent collaboration and Flows for event-driven control	MIT
LangGraph	Orchestration framework	"Low-level orchestration framework for building stateful agents"	MIT
MetaGPT	Multi-agent framework	Takes a one-line requirement and outputs user stories, requirements, data structures, APIs, and documents using product manager, architect, project manager, and engineer roles	MIT

The three coding agents work on code directly. CrewAI and LangGraph are general-purpose frameworks, so you write the coding workflow yourself. SWE-agent's README adds that most of its current development effort is on mini-SWE-agent, "which has superseded SWE-agent," and recommends mini-SWE-agent going forward.

How do the frameworks differ?

CrewAI's README describes Crews, role-based agents that collaborate autonomously, and Flows, which give event-driven control over a workflow. LangGraph's README lists durable execution, human-in-the-loop oversight, and short-term and long-term memory for stateful agents. MetaGPT's README describes a software-company workflow built from LLM roles. With CrewAI and LangGraph, expect to write the coding workflow and its tools yourself.

How does Databricks support open-source agents?

Databricks documents two paths that apply to open-source agents:

• Govern a coding agent's models. The Unity Gateway CLI (ug) connects coding agents such as Claude Code and Codex, and the open-source OpenCode, to models, MCP tools, and skills through Unity Gateway. A tutorial shows how Unity Catalog controls who can use which models while Unity Gateway enforces rate limits and records usage.
• Host agents you build. Agent Bricks is the Databricks agent developer platform. Agent Runtime hosts agents built with any framework or harness, and the Agent Bricks CLI scaffolds LangGraph projects. Agents can run the code they write in Databricks Sandbox, an isolated environment, and MLflow Tracing records each step an agent takes.

FAQs

Can an open-source coding agent use my own models?

OpenHands says you can use it with any LLM, and SWE-agent says it works with the language model of your choice. For OpenCode, Databricks documents a manual configuration that points it at Unity Gateway model services.

Is SWE-agent still the project to start with?

Its README says development effort is concentrated on mini-SWE-agent and recommends using mini-SWE-agent instead of SWE-agent going forward. Read both READMEs before you choose.

Do I need CrewAI, LangGraph, or MetaGPT to replace Devin?

Not necessarily. CrewAI and LangGraph provide building blocks for multi-agent or stateful workflows rather than a finished coding agent. If you want an agent that already works on repositories, look at OpenHands, SWE-agent, or OpenCode first.

How should I compare an open-source agent with Devin?

Run each candidate on the same set of your own repositories and tasks, and review results you can inspect. This page does not rank the projects or publish benchmark results.