Open-Source on Vibe Coding

Cline

Mon, 01 Jan 0001 00:00:00 +0000

Cline is an open-source VS Code extension that acts as an autonomous agent — creating and editing files, running terminal commands, and even using a browser — while showing exactly what it plans to do before it does it.

GitHub Topic: AI Agents

Mon, 01 Jan 0001 00:00:00 +0000

GitHub’s topic pages aggregate repositories tagged with a given subject — in this case, AI agent frameworks and projects — sorted by popularity and updated automatically as the ecosystem moves.

Aider

Mon, 01 Jan 0001 00:00:00 +0000

Aider was one of the first tools to popularize the “AI edits your local repo from the terminal” workflow. It reads your codebase, makes changes as real git diffs, and commits each one with an auto-generated message, so every edit is reviewable and revertible.

Codex CLI

Mon, 01 Jan 0001 00:00:00 +0000

Codex CLI is OpenAI’s terminal-based coding agent: an open-source command-line tool that reads a local repository, plans a multi-step change, edits files, and runs commands, with approval settings ranging from step-by-step confirmation to fully autonomous runs.

Gemini CLI

Mon, 01 Jan 0001 00:00:00 +0000

Gemini CLI is Google’s open-source command-line agent built on Gemini models: it works directly against a local repository from the terminal, answering questions about the code, making edits, and running commands as part of a multi-step task.

Zed

Mon, 01 Jan 0001 00:00:00 +0000

Zed is a code editor built from the ground up (not forked from VS Code) for raw editing and rendering speed, with AI collaboration — chat, inline edits, and agentic multi-step changes — designed in as a core feature rather than bolted on through extensions.

Continue.dev

Mon, 01 Jan 0001 00:00:00 +0000

Continue.dev is an open-source extension that adds chat, autocomplete, and agentic editing to VS Code or a JetBrains IDE, with every model provider — hosted or fully local — configurable through one JSON config file rather than being tied to a single vendor.

OpenHands

Mon, 01 Jan 0001 00:00:00 +0000

OpenHands (formerly OpenDevin) is an open-source platform for running autonomous coding agents that execute real engineering work — planning a task, writing code, running commands, and applying changes — rather than just suggesting edits, with a composable SDK for building custom agent workflows on top of it.

Chef

Mon, 01 Jan 0001 00:00:00 +0000

Chef is Convex’s prompt-to-app builder, distinguished from most “vibecoding” app generators by actually understanding backend concerns: it scaffolds a real Convex database schema, server functions, and authentication alongside the frontend, rather than generating UI backed by mock data.

Void

Mon, 01 Jan 0001 00:00:00 +0000

Void is an open-source fork of VS Code built as a transparent alternative to Cursor and Copilot: every AI request is visible and configurable, and nothing routes through a proprietary backend unless you choose to.

Pear AI

Mon, 01 Jan 0001 00:00:00 +0000

Pear AI is an open-source code editor built on top of VS Code and the Continue extension, aiming to package the best available AI coding tools into a single, ready-to-use interface rather than making users assemble their own stack of extensions.

Melty

Mon, 01 Jan 0001 00:00:00 +0000

Melty is an open-source code editor built around a persistent chat pane rather than inline suggestions: it watches your terminal, compiler output, debugger, and GitHub activity so it can propose changes, fix errors, and draft commit messages with real context on what you’re doing.

Tabby

Mon, 01 Jan 0001 00:00:00 +0000

Tabby is an open-source coding assistant you run on your own infrastructure, giving teams Copilot-style completions and chat while keeping code entirely inside their own network instead of a vendor’s cloud.

Refact.ai

Mon, 01 Jan 0001 00:00:00 +0000

Refact.ai is an open-source coding assistant that pairs unlimited code completions with an in-IDE chat and agent for refactoring and multi-file changes, backed by a codebase-aware vector index and available either as a managed plan or a self-hosted deployment.

Aide

Mon, 01 Jan 0001 00:00:00 +0000

Aide is an open-source VS Code fork from CodeStory AI that merges chat and editing into one flow — you can brainstorm a problem by referencing files in chat, then jump straight into applying edits across multiple files in the same session.

OpenCode

Mon, 01 Jan 0001 00:00:00 +0000

OpenCode brings an agentic coding workflow into the terminal with a polished text-user-interface, built provider-agnostic so it can run against Anthropic, OpenAI, local models, or anything else you point it at.

Goose

Mon, 01 Jan 0001 00:00:00 +0000

Goose, built by Block (the company behind Square and Cash App), is an open-source agent that runs on your own machine and can plan, write code, execute commands, and iterate on a task with minimal hand-holding.

Plandex

Mon, 01 Jan 0001 00:00:00 +0000

Plandex is a terminal AI coding agent aimed at tasks too large for a single chat exchange — refactors and features spanning dozens of files — with a sandboxed diff review step before changes are applied.

Crush

Mon, 01 Jan 0001 00:00:00 +0000

Crush is Charm’s entry into the terminal coding agent space, pairing their signature polished TUI design with an agent that can read, edit, and run code across your project, and switch between LLM providers on the fly.

RA.Aid

Mon, 01 Jan 0001 00:00:00 +0000

RA.Aid (Research-Augmented AI Development) is an open-source CLI agent built around a research-then-plan-then-execute loop, so the agent gathers context on how your codebase actually works before proposing a change.

Why it stands out

Explicit research phase. Instead of jumping straight to edits, RA.Aid first investigates relevant code paths and dependencies.
Separates planning from execution, making it easier to review the intended approach before code gets touched.
Open source and self-hostable, with no dependency on a proprietary backend.

Good for

Developers who want an agent that reasons through unfamiliar code before making changes rather than editing on the first pass.

Qwen Code

Mon, 01 Jan 0001 00:00:00 +0000

Qwen Code is Alibaba’s open-source command-line coding agent, built as a fork of Google’s Gemini CLI and adapted to work best with Qwen models while still supporting other providers.

Why it stands out

Purpose-tuned for Qwen models, with prompts and tool-calling behavior optimized for that model family’s strengths.
Built on a proven base. Forking Gemini CLI means it inherits a mature agentic loop rather than starting from scratch.
Fully open source, so it can be self-hosted and modified freely.

Good for

Teams standardizing on Qwen models who want a terminal agent tuned specifically for them.

Smol Developer

Mon, 01 Jan 0001 00:00:00 +0000

Smol Developer is one of the earliest open-source “AI junior developer” tools: point it at a plain-text spec describing what you want built, and it scaffolds a working small application from scratch.

SWE-agent

Mon, 01 Jan 0001 00:00:00 +0000

SWE-agent wraps an LLM in an “Agent-Computer Interface” — a constrained set of commands for viewing, searching, and editing files and running tests — so it can work through real GitHub issues end-to-end. It’s the reference agent behind much of the SWE-bench benchmark work.

Devika

Mon, 01 Jan 0001 00:00:00 +0000

Devika emerged as an open-source answer to Devin — an agent that takes a high-level coding objective, plans the steps to get there, browses for information it needs, and writes and iterates on code with visibility into its reasoning at each step.

MetaGPT

Mon, 01 Jan 0001 00:00:00 +0000

MetaGPT assigns different LLM agents distinct roles — product manager, architect, project manager, engineer — and has them collaborate through standard software-team artifacts like PRDs and design docs on the way to generating a working codebase from a single requirement.

bolt.diy

Mon, 01 Jan 0001 00:00:00 +0000

bolt.diy takes StackBlitz’s Bolt.new — an AI app builder that runs a full dev environment in the browser via WebContainers — and opens it up so anyone can plug in their own model, including local or open-weight ones, instead of being locked to a single provider.

Onlook

Mon, 01 Jan 0001 00:00:00 +0000

Onlook works the opposite direction from most design-to-code tools: it opens an existing React and Tailwind app in a visual, Figma-like canvas so designers can edit a real running product directly, with changes written back into the actual source code.

Screenshot to Code

Mon, 01 Jan 0001 00:00:00 +0000

Screenshot to Code takes a single image — a screenshot, a hand-drawn mockup, or a Figma export — and uses a vision-capable AI model to reconstruct it as working front-end code, with no design file required.