# TryCase TryCaseは、GitHubのプルリクエスト向けAI支援型エンドツーエンドテストツールです。使い捨てのLinux環境で影響を受けるユーザーフローを実行し、判定結果、動画、スクリーンショット、失敗の詳細をプルリクエストに投稿します。 - Findings24: - ウェブサイト: - 料金: Freemium - カテゴリー: Video Tools ## TryCaseとは？ TryCase tests pull requests by reading the code changes, selecting the user journeys they affect, and exercising those journeys in a disposable Linux environment. Web applications are tested in a browser, while APIs, CLIs, workers, and packages are exercised through their real commands. Each run produces a verdict and supporting evidence on GitHub, including a captioned video and final screenshot. TryCase is designed to provide pull-request feedback without requiring teams to write or maintain a browser test suite. ## TryCaseでできること ### Pull-request-triggered verification Connect selected public or private GitHub repositories and run a check for each pull request, either on every push or when requested with a \`/trycase run\` comment. ### Journey selection from code changes TryCase reads the pull request and determines which user journeys or changed flows to test, rather than requiring a separately authored browser test for each run. ### Browser and command-based testing The service opens web applications in a browser and works through them like a user; APIs, CLIs, workers, and packages are exercised through their real commands. Monorepos are supported through repository setup that discovers the applications once and replays the setup for later runs. ### Evidence attached to GitHub Results include a pass or fail verdict, a captioned run video, and a final screenshot. Failure comments explain what was expected and what actually happened. ### Bring-your-own AI or use TryCase Auto Users can connect a ChatGPT subscription through Codex or provide an OpenRouter API key and supported model. TryCase Auto runs the model connection on the service side; Claude subscription support is listed as coming soon. ### CLI and controlled secret handling The Node 20+ CLI can start, follow, continue, stop, and inspect verifications, and can list or download artifacts. Requested development or test secrets are sent by approved names from an environment or file rather than as command-line values. ## 利用シーン ### Reviewing interface changes before merge A team can use a pull-request run to check a changed screen or user flow and review the recorded walkthrough alongside the GitHub verdict before deciding what ships. ### Testing repositories without a browser suite Projects that do not maintain end-to-end browser tests can have TryCase infer affected journeys from the pull request and exercise the live application instead of first creating a test suite. ### Verifying non-browser software Teams working on APIs, command-line tools, workers, or packages can use the same verification workflow through real commands, provided the software and setup run on Linux. ### Checking large or multi-application repositories Monorepo teams can configure application setup once and have later runs replay it while the agent focuses on the pull request’s affected targets. ### Running verification from coding-agent workflows Developers and automation can use the CLI to start a pull-request verification, stream status as NDJSON, respond to attention requests, and retrieve saved screenshots or captioned videos. ## よくある質問 ### Does TryCase require prewritten end-to-end tests? No. TryCase reads the pull request, identifies the user journeys affected by the changes, and attempts those journeys. It is intended to avoid writing and maintaining a separate browser test suite for this workflow. ### How does TryCase connect to GitHub? Install the TryCase GitHub App on selected public or private repositories. Each pull request then receives its own check and a result comment. The trigger can be every push or an explicit \`/trycase run\` comment. ### What evidence does a run produce? A completed run returns a pass or fail verdict, a captioned video of the run, and a final screenshot. If a check fails, the pull-request comment describes the expected and actual behavior. ### Can TryCase test software other than web applications? Yes, the site says it can test anything that runs on Linux. Web interfaces are exercised in a browser, while APIs, CLIs, workers, and packages are tested through their real commands. ### Where does a pull-request run execute? Each pull request receives its own disposable environment with 4 vCPUs and 8 GB of memory, without swap. The environment is deleted when the run ends, and the site recommends using non-production values. ## クイック情報 - **Category**: Developer Tool; AI Testing - **Primary integration**: GitHub pull requests - **Supported execution**: Linux environments; browser and command-based testing - **CLI requirement**: Node 20 or later - **Run environment**: Disposable environment with 4 vCPU and 8 GB memory - **Pricing model**: Pay per pull request with free monthly Small or Medium runs and prepaid balance