
Aider
代替候補のひとつ
AI ターミナルでのペアプログラミング — お気に入りのLLMでコードベースを編集しましょう。
An open-source coding agent for the terminal, desktop, and IDE with 75+ model providers
OpenCode is an open-source AI coding agent for the terminal, desktop, and IDE. It runs multiple sessions in parallel, loads the right LSPs automatically, supports 75+ model providers and local models, and can share session links for debugging or review.
更新日
2026-10-21
対応プラットフォーム
Windows / macOS / Linux
お試し
Freemium: OpenCode is open source and free to install, with free models included and support for your own provider API keys, including GitHub Copilot and ChatGPT Plus/Pro accounts. OpenCode Zen is an optional pay-as-you-go model gateway with a $20 starting balance, zero markups, and automatic top-up; confirm current model prices and limits on the official site.
* 料金は参考情報です。最新の料金は公式サイトをご確認ください
Use the same open-source agent in the terminal, desktop app, or IDE extension.
Start multiple agents on the same project and switch between them without losing context.
Connect providers through Models.dev, or use GitHub Copilot and ChatGPT Plus/Pro accounts.
Share a link to any session and let OpenCode load the right language servers automatically.
Run an open-source agent locally with your own providers; the site states it does not store code or context data.
Keep parallel sessions, LSP context, and coding tasks in one terminal workflow.
Reuse GitHub Copilot or ChatGPT Plus/Pro accounts, or connect 75+ providers instead of buying another model plan.
クイックスタート
Install the agent and start it inside one repository, then connect a provider you already pay for, such as GitHub Copilot or your own API key, before writing a prompt.
Open a second session for the same project instead of switching tasks in one thread: keep exploration in one session and the actual edit in another.
Let it read files through the LSP-backed context first, then ask for the change. Accurate symbols come from the language server, not from pasting file contents.
実践のコツ
Parallel sessions are the real advantage and also the easiest way to waste tokens. Give each session one goal and one directory, and start a fresh session when you switch tasks instead of stacking context. When a provider feels weak, switch models inside the same session before rewriting the prompt, because the surrounding context usually matters more than the model tier.
このツールが合わない場合は、次の代替候補をご覧ください