qmd-search

qmd-search turns AI coding session logs from Claude Code, Codex, and Pi into searchable Markdown for QMD, with a practical playbook for syncing and indexing them on a remote GPU server.

Minimal exporters and a deployment playbook for indexing local AI session logs with QMD.

Current scope:

• Codex JSONL -> Markdown exporter
• Pi JSONL -> Markdown exporter
• playbook for:
  • exporting sessions on a laptop
  • installing Node + QMD on a remote GPU server with mise
  • syncing markdown to a remote server
  • indexing and querying with QMD

For Claude Code, this repo currently recommends using the existing external exporter:

• uvx claude-code-log@latest

Repo contents

• scripts/exporters/codex_jsonl_to_md.py
• scripts/exporters/pi_jsonl_to_md.py
• docs/index-ai-coding-session-logs-with-qmd-on-a-remote-gpu-server.md

Quick start

Codex

python3 scripts/exporters/codex_jsonl_to_md.py \
  ~/.codex/sessions/YYYY/MM/DD/rollout-<session-id>.jsonl \
  ~/qmd-sessions/main-laptop/codex/<session-id>.md

Pi

python3 scripts/exporters/pi_jsonl_to_md.py \
  ~/.pi/agent/sessions/<encoded-cwd>/<timestamp>_<session-id>.jsonl \
  ~/qmd-sessions/main-laptop/pi/<session-id>.md

Claude Code

uvx claude-code-log@latest \
  ~/.claude/projects/<encoded-project>/<session-id>.jsonl \
  -f md \
  -o ~/qmd-sessions/main-laptop/claude-code/<session-id>.md

Then follow docs/index-ai-coding-session-logs-with-qmd-on-a-remote-gpu-server.md.

Notes

• QMD can index raw JSONL, but Markdown works better.
• The current Codex and Pi exporters are intentionally small first-pass converters.
• Claude Code main-session files are the best first source. Excluding subagents/ initially is recommended.

License

MIT