Both Aider and Continue.dev are open-source, bring-your-own-model AI coding tools. Neither locks you into a single LLM, neither charges a subscription for the software itself, and both have been around long enough to feel stable rather than experimental. The thing that actually separates them is where you sit while you work: Aider lives in your terminal and treats your git repo as the unit of work; Continue.dev lives inside VS Code or JetBrains and treats your open editor buffer as the unit of work.
We ran both against the same small TypeScript project for a week — the same feature requests, the same models (Claude and a local model through Ollama) — to see where the terminal-first and editor-first philosophies diverge in practice, not in marketing copy.
How each one wants you to work
Aider is a command-line program. You launch it inside a git repository, point it at the files you want to change, and talk to it in a REPL. When it edits, it writes the change directly to disk and — by default — makes a git commit for every edit it applies. That last detail is the whole personality of the tool. Aider assumes your repository is the source of truth and that every AI change should be a reviewable, revertable commit. If a change is wrong, you git diff it or /undo it, and the bad commit is gone.
Its other defining feature is the repository map. Rather than dumping your whole codebase into the prompt, Aider builds a compressed map of your symbols and file structure and sends the model just enough to reason about what it can't see. That keeps token usage down on larger repos and is the main reason it stays usable in projects with hundreds of files.
Continue.dev is an extension, not a program. You install it into VS Code or a JetBrains IDE and it adds three things to the editor you already use: inline autocomplete as you type, a chat sidebar that can see your open files and highlighted selections, and an edit/agent mode that applies changes to your buffers. Context comes from what you give it — the active file, a selection, or @-references to other files, docs, or the terminal output. Nothing gets committed automatically; changes show up as a normal editor diff you accept or reject inline.
This is not a closed-vs-open or cheap-vs-expensive comparison. Both tools are MIT-licensed and model-agnostic — you can run either against Claude, a local Ollama model, or anything with an OpenAI-compatible endpoint. The cost you pay is the model's API cost, which is identical whichever tool sends the request. Pick based on workflow, not licensing.
Where the difference actually bites
The split shows up the moment a change touches more than one file. Aider's auto-commit-per-edit means a three-file refactor lands as a clean sequence of commits you can read, bisect, or roll back individually. When the model went sideways on a rename, reverting was one command and the working tree was clean again. There is no "accept all these scattered diffs" step — the diffs are commits, and git is the review surface.
Continue.dev keeps you in the editor's review loop instead. You see each proposed hunk in the gutter and accept or reject it in place, which is faster for single-file work because you never leave the file you were already reading. The cost is that multi-file changes are less ceremonial: you are accepting hunks across tabs, and your git history reflects whatever you decide to stage afterward, not the AI's step-by-step reasoning.
Autocomplete is the cleanest functional gap. Continue.dev ships a real fill-in-the-middle autocomplete provider — the grey-text suggestions you tab to accept while typing. Aider has nothing equivalent; it is a conversational tool, not a typing assistant. If "AI finishes my line as I type" is a workflow you rely on, that alone decides it.
Context control runs the other way. In Aider you explicitly /add files to the chat, so you always know exactly what the model can see, and the repo map fills the gaps. In Continue.dev context is more implicit — the open file plus whatever you @-mention — which is lower-friction for quick questions but easier to get wrong on a large change, because the model may be reasoning about less than you assume.
Because both are free and model-agnostic, plenty of developers run them together: Continue.dev for autocomplete and quick in-editor questions, Aider for larger, git-tracked refactors driven from the terminal. They share the same API key and step on nothing. If you are undecided, install both for a week before committing to one habit.
Which one fits you
Reach for Aider if you live in the terminal, care about a clean and auditable git history, and do work that spans multiple files. The auto-commit model and repo map are built for exactly that: large changes you want to review as commits and revert surgically when the model is wrong. It rewards developers who already think in git diff.
Reach for Continue.dev if your center of gravity is the editor, you want autocomplete in the loop, and most of your AI use is quick, single-file edits and questions about code you are already looking at. The inline review flow keeps you in one window, which is the faster feedback loop for that style of work.
If you want the most polished version of the editor-first experience and are willing to pay for a managed product rather than wiring up an open-source extension, a dedicated AI-native editor is worth a look alongside Continue.dev.
The honest summary: this is a workflow choice, not a capability ranking. The same Claude or local model does the actual thinking in both. What you are picking is whether your AI assistant should meet you in the terminal, where the repository is the contract, or in the editor, where the open buffer is.
Originally published at pickuma.com. Subscribe to the RSS or follow @pickuma.bsky.social for new reviews.
Top comments (0)