title: Why dkod description: The problem with vanishing agent history and how git-native capture solves it.

Why dkod

The problem

AI agents write code every day. Claude Code, Codex, Cursor, Gemini CLI — your team runs them constantly. But the moment a session ends, the history disappears:

• No record of what the agent did. The transcript — every prompt, every response, every tool call — is gone.
• No record of why. The reasoning behind code changes vanishes. When something breaks, you're left reverse-engineering diffs.
• No cross-repo visibility. Each developer's agent history is siloed on their machine. The team has no way to search or learn from past sessions.
• No accountability. AI-authored code enters the codebase with no provenance trail.

The solution

dkod captures every agent session and stores it as a git ref inside your repository.

Git-native. Sessions are git blobs under refs/dkod/sessions/. They push and fetch with standard git push and git fetch. No external database, no cloud dependency, no new infrastructure.

Privacy-first. Transcripts never leave your git host. The team indexer stores metadata and embeddings only — full content is always fetched live through the user's GitHub token.

Redaction by default. Built-in patterns strip AWS keys, GitHub tokens, OpenAI keys, Stripe keys, and env assignments before anything is stored. Over-eager redaction is fixable; leaked credentials in git history are not.

Agent-agnostic. One tool for all your agents. No vendor lock-in.

Solo vs. team

Next steps

• Quickstart — install and capture your first session
• Privacy Model — the full trust boundary explanation