opencode

Aider vs opencode: The Open-Source Terminal Agents Compared

Crazy Black Dog

07 Sep 2026 • 4 min read

For developers who live in the terminal, choosing an AI coding assistant used to mean choosing between closed, proprietary products — or doing without. That is no longer true. Two open-source projects now dominate the conversation: Aider, the veteran pair-programmer that taught a generation of developers to code with git, and opencode, the newer, more ambitious agent built by the SST team (now called Anomaly). Both run in your terminal, both are free and open-source, and both are genuinely excellent. Yet they embody two very different ideas of what an AI coding tool should be. This article compares them head to head, so you can decide which one belongs in your daily workflow.

The Quick Comparison

Here is the short version before we go into details:

	Aider	opencode
Tagline	AI pair programming in your terminal	Open-source AI coding agent for the terminal
Author / ecosystem	Started by Paul Gauthier, now `Aider-AI/aider` on GitHub	Developed by the SST team (now Anomaly) — `anomalyco/opencode`
Language	Python	TypeScript
Models supported	A very broad range, including open-weight models	Multi-provider: frontier APIs down to local models
Core workflow	Git-centric: proposes patches, commits automatically	Agentic: executes tasks end to end
Extensibility	Model-agnostic and scriptable, with a huge install base	Sessions, plugins, and MCP support
Philosophy	A pair programmer you review	An autonomous agent you supervise

Aider: The Granddaddy of Terminal Pair Programming

Aider describes itself, simply and accurately, as "AI pair programming in your terminal." Created by Paul Gauthier and released in 2023, it is the tool that made terminal-based AI coding mainstream. It is now maintained under the Aider-AI/aider umbrella on GitHub, with an enormous footprint: around 6.8 million installations on PyPI, and roughly 15 billion tokens processed every week by its users — numbers that make it one of the most-used AI coding tools in existence.

What makes Aider distinctive is how deeply it is wired into git. When you ask it to change something, it does not just print a suggestion: it reads your repository, applies edits to the relevant files as patches, and commits each change with a descriptive, well-formatted commit message. Every modification is a checkpoint you can inspect, revert, or squash. For teams that live by git log, this is transformative — the AI's work becomes part of your normal review flow instead of a black box.

Aider is also famously model-agnostic. It supports a very large range of models — frontier models, but also many open-weight options — and it works with OpenRouter, the neutral marketplace where many of those open-weight models are served. If you care about running your assistant on open models, Aider has been there from the start.

Perhaps the best proof of the project's maturity is its own "singularity" metric: about 88% of the new code in the latest Aider release was written by Aider itself. The dogfooding is real.

opencode: The New-Generation Agent

opencode is the younger contender, built in TypeScript by the SST team (now called Anomaly) (the anomalyco/opencode repository on GitHub). It arrived with a different ambition: not just to pair-program with you, but to act as a complete agent. It is one of the most-starred AI coding agents on GitHub, and it shows what a second-generation terminal tool looks like.

Where Aider is a chat loop around git, opencode is a full environment. It organizes your work into sessions you can resume, ships a plugin system for extending its behavior, and supports the Model Context Protocol (MCP), which lets it reach into external tools and data sources. It is multi-provider by design: you can point it at frontier APIs, at any provider you already use, or at local models running on your own machine — and switch between them without changing your workflow.

The bigger difference is autonomy. opencode is built to take a task and carry it through: planning, editing files, running commands, and iterating on results until the job is done. You supervise, steer, and intervene, but the agent does the walking. For larger, multi-step tasks — refactors, migrations, wiring up a new feature across several files — that style of working feels closer to delegating to a junior engineer than to typing prompts at a REPL.

Two Philosophies of Terminal AI

Strip away the features and the real difference is philosophical. Aider's model is the pair programmer: it proposes changes, commits them in small reviewable steps, and expects you to stay in the loop. Its git-native design is not an implementation detail — it is the whole point. Every edit is a patch you can read; every commit is a decision point. Control never leaves your hands.

opencode's model is the agent: give it an objective, and it executes. Sessions, plugins, and MCP exist to make that execution smooth and connected to the rest of your tooling. It is less about reviewing each patch and more about steering an autonomous process toward a goal. That is more powerful, and also more demanding: you need to trust — or at least verify — what the agent did, because it does a lot more on its own.

Neither approach is objectively better. Git-centric and conservative suits certain workflows; session-based and autonomous suits others. The right question is not "which is more capable" but "which matches how you like to work."

Which One Should You Choose?

If you are new to AI-assisted development, or if your workflow is built around git — pull requests, code review, clean history — Aider is the natural starting point. Its commit-per-change flow teaches you what the AI is doing, keeps you in control, and works beautifully with open-weight models if that matters to you.

If you are a power user juggling many providers, or you want an assistant that can carry a multi-step task from start to finish with plugins and MCP, opencode will feel like a leap forward. It is the better fit for a full workflow: sessions to come back to, an ecosystem to extend, and an agent that actually gets things done.

And the honest verdict? You cannot really go wrong. Both tools are free, open-source, and excellent. Aider remains the safest, most proven choice — years of battle-tested development, wrapped in a git-native workflow that has scaled to millions of installs. opencode is the more exciting one, pushing the agent paradigm forward with a modern, extensible design. Many developers end up with both installed: Aider for surgical, reviewable edits, opencode when a task needs an agent. Start with the one that matches your philosophy — and keep the other in your back pocket.