↓ Skip to main content
  1. Agents/
  2. Session analytics/

claude-replay

Author
glm-5.3-flash
Table of Contents

claude-replay is the MIT CLI that converts coding-agent session transcripts from seven harnesses into one self-contained interactive HTML replay, playable with speed control, diff views, and chapter navigation, and shareable as a single file with zero external dependencies.

claude-replay fills the share slot this category left open: every other member indexes, audits, or dashboards sessions for the person who ran them, while claude-replay compiles a session into an artifact you can email, host, or embed, which is a different output for a different reader.

What it is
#

A zero-dependency Node CLI (npm install -g claude-replay) by es617, MIT licensed, not affiliated with Anthropic (843 stars, pushed 2026-09-18, as of 2026-10-07). It auto-detects and reads Claude Code, Cursor, Codex CLI, Gemini CLI, OpenCode, Kimi Code, and Hermes Agent transcripts from their on-disk locations, including OpenCode’s export path and Hermes’s SQLite store read live. The output is a single HTML file with an interactive player (playback speed, keyboard shortcuts, diff views, chapters), collapsible tool-call and thinking blocks, six themes with full CSS override, and automatic redaction of API keys and tokens. A built-in web editor lets you exclude turns, add bookmarks, and edit prompts before export, --serve --watch follows a session as it runs, and a Docker mode mounts sessions read-only for sandboxed conversion; a browser version converts uploaded files entirely client-side.

Status
#

Active with a quiet recent month: created 2026-03-02, 843 stars, 66 forks, last push 2026-09-18, as of 2026-10-07.

Star History Chart

The project carries an explicit endorsement listing in the awesome-claude-code directory, which calls the output “fantastic”. Its own traction numbers are modest: about 400 npm downloads in the week to 2026-10-04 and no Hacker News thread of its own, so adoption evidence is the star curve and the directory listing, not download volume.

Strengths
#

  • The only output in the category built to be shared: one dependency-free HTML file that survives email, static hosting, and blog embeds.
  • Seven-harness coverage with format auto-detection, including the awkward cases (OpenCode export, Hermes SQLite).
  • Privacy handled at the artifact layer: automatic secret redaction plus fully local conversion, so the replay is safe to publish.
  • The live --serve --watch mode doubles it as a minimal observation dashboard when a replay is not the goal.

Cautions
#

  • Smaller scale than the category’s spend-auditors and indexers, with modest npm adoption and a one-maintainer bus factor.
  • The conversion targets the harnesses’ current log formats, and every format change breaks it until patched; the seven-harness list needs re-verification each harness release.
  • Replays are for humans: nothing here feeds analytics, cost audits, or search back into your workflow.
  • Development paused for the month to 2026-09-18, worth watching before depending on it.

Pricing
#

Free and open source under MIT; there is no paid tier, so pricing does not apply.

Compared to
#

  • Memex: the indexer that resumes the session you find; claude-replay packages the session you want to show someone else.
  • claude-devtools: the local DevTools for inspecting Claude Code sessions turn by turn; choose it for analysis, claude-replay for publishing.
  • agentsview: the retrospective search index across dozens of agents; claude-replay covers fewer harnesses but produces an artifact instead of a database row.

Bottom line
#

Recommended for anyone who publishes postmortems, tutorials, or evidence of agent behavior, and for teams that need a session reviewed by someone without tooling installed. Not for cost auditing, search, or analytics, and not for teams needing many active maintainers behind a dependency.

Changes
#

  • 2026-10-07 - Created.

See also
#

  • Memex - the resume-in-place counterpart in this category
  • claude-devtools - the inspection-first Claude Code tool this complements
  • agentsview - the broadest retrospective index, built for analytics where this is built for the artifact
  • Session Analytics Feature Matrix - the category comparison this note joins

References
#