Line-level explanations with receipts
Pruve answers questions about specific lines by tracing them back to the changes and discussions that wrote them. The result is a clickable chain of evidence rather than a speculative summary.
Pruveはリポジトリの履歴を読み込み、特定のコード行がなぜそのように書かれているのかを、開いて確認できる根拠とともに説明します。GitHubまたはGitLabのリポジトリから、開発者、プロダクトマネージャー、AIエージェント向けに読み取り専用のコンテキストを提供します。
Pruve is a read-only product for exploring why code exists the way it does. It reads a repository’s history and ties individual lines back to the change that introduced them, so you can open the supporting discussion, commit, review thread, or agent session instead of relying on guesswork.
The product is built around evidence. When a line can be explained, Pruve shows the receipts behind it; when it cannot, it says so plainly. That makes it useful for understanding unusual constants, merge decisions, undocumented behavior, and other code that is hard to interpret from the source alone.
Pruve answers questions about specific lines by tracing them back to the changes and discussions that wrote them. The result is a clickable chain of evidence rather than a speculative summary.
It classifies changes as explained, described, or undocumented, then surfaces that coverage in dashboards, file trees, and the code gutter so readers can see which parts of a codebase have recorded reasoning.
Indexing can pull from commit messages, pull request bodies, review comments, linked issues, and agent sessions, which lets the product preserve the discussion around a change, not just the diff.
Pruve reads repository history and file contents without writing back to the code host. The only optional write action mentioned is a PR check run for the documentation gate.
The product exposes read-only MCP tools so agents can inspect reasoning before editing code. This is meant to let automated workflows ask “why” before making a change.
Teams can optionally fail pull requests that have no recorded reasoning anywhere, turning missing context into an explicit review signal.
A developer can click a line, see the change that introduced it, and read the discussion that explains why it was added instead of reverse-engineering the intent from the implementation.
When a merge commit drops discussion context, Pruve can surface the issue, benchmark, review comment, or agent session that explains why a choice was made.
Teams can scan a codebase to see which files or folders have recorded reasoning and which areas are still undocumented, making gaps visible before they become maintenance problems.
An AI agent can use MCP tools to inspect receipts before editing code, helping it avoid changes based on invented rationale.
Teams that want explicit evidence for pull requests can use the documentation gate to flag changes that lack recorded reasoning.
No. The product is read-only. The documentation says it does not push code, comments, or branches. The only write-related action mentioned is an optional PR check run for the documentation gate.
It indexes repository history and related discussion, including commits, pull requests, review threads, issues, and agent sessions when they are available.
It says so rather than guessing. The product is designed to report what it searched and stop if there is no evidence.
Coverage is counted as explained, described, or undocumented changes. Those counts are used in the product’s dashboards and line gutters.
Yes. The pricing page says open source is free forever, with free and unmetered read-only access for receipts.
トラフィックデータは参考情報としてご利用ください。
refraction.dev
リファクタリング、ドキュメント、テスト、バグ検出、コード変換に対応するAIコード生成ツール
onorca.dev
Orcaはコーディングエージェントで開発を進めるためのエージェント開発環境です。分離されたworktreeで複数のCLIエージェントを並列実行し、デスクトップとモバイルの連携ワークフローに対応します。
blopai.com
ブラウザテストを作成・実行・分類・修正するQAエージェント
rlama.dev
RAGシステムとAIエージェントを構築するローカルAIプラットフォーム
kastra.ai
Kastraは、プロンプト、ツール呼び出し、シェルコマンド、APIリクエスト、ブラウザ操作を実行前に検査する、AIシステム向けの認可インフラです。ポリシー適用、署名付き監査証跡の保持、ローカルおよびエンタープライズAIワークフローのガバナンスを支援します。
aakar-ai.dev
AIプロバイダー、プロジェクト設定、ログ、分析を1つのダッシュボードで管理。BYOK対応。