Stop re-explaining your codebase to your coding agent

CLAUDE.md and AGENTS.md hold what you wrote down before the work started. They do not update themselves when a run makes a decision, discovers a constraint or leaves something unfinished. Runtive records what the run produced — decisions, current state and the recommended next step — extracted from the session and the diff rather than from anyone's memory of them, and keeps it in one context file in your repository for Claude Code and Codex alike.

Pricing

Runtive Workspace — $19 USD / month, with a 14-day free trial. Billed monthly, cancel anytime. See what is included.

How it works

• Run runtive init once per repository. It adds a reference to .runtive/context.md to the instruction files your agents read — CLAUDE.md, AGENTS.md, or both — and leaves a file that already has its own instructions about it as it is.
• Finish a task with runtive finish.
• Runtive finds the Claude Code or Codex session on its own and captures the transcript and the git diff. Point it at your checks with --checks or --checks-cmd and it captures their output too.
• It extracts the decisions, current state and recommended next step, and refreshes .runtive/context.md with them. Each entry traces back to the run it came from.

Your model credentials

Runtive does not resell inference and never bills you for tokens.

• Runtive chooses the extraction path automatically. Today the local path it reaches for is Claude Code.
• Local extraction runs against the login already on your machine and sends no model credential to Runtive's backend.
• Remote API — used when you configure an Anthropic API key for Runtive, or ask for it explicitly — sends the key with that single request, and it is not stored on our servers.

Contact

Support, security and privacy: support@runtive.dev