codex-claude-transfer

Move local Codex and Claude Code sessions between machines. The command is cct.

Unofficial. Not affiliated with or endorsed by OpenAI or Anthropic. These tools' internals can change at any time and break this tool. Use at your own risk. See the Disclaimer.

Codex and Claude Code keep valuable project context in local session files. That is good for privacy, but painful when you switch machines, rebuild a laptop, or want to continue a thread in the other agent.

cct turns that local context into a simple handoff:

Machine A:  cct export --project .      ->  project.codexbundle
                        copy it however you trust
Machine B:  cct import ./project.codexbundle

No cloud account, hosted sync service, or background process by default. You move one .codexbundle over a channel you trust, then import it after checksum verification. The agents' indexes are left alone and re-scan the session files themselves.

It works for Codex, Claude Code, and cross-agent handoff. The same flow is available as a CLI, a terminal wizard (cct ui), and a local browser app (cct app).

Export on one machine, then import or incrementally sync onto the other:

The optional local desktop UI uses the same engine:

More clips in demo/: LAN sync, full-text search, secret scan & redact, Markdown export, repair-times, cross-agent handoff, encryption, conflict resolution & cwd remap, export filters, git handoff, the interactive cct ui wizard, and reading compressed .jsonl.zst sessions.

All recordings use throwaway demo sessions, never a real ~/.codex or ~/.claude.

# From source (Go 1.23+)
go install github.com/ahmojo/codex-claude-transfer/cmd/cct@latest

Or download a prebuilt binary from Releases, or build from a clone:

git clone https://github.com/ahmojo/codex-claude-transfer.git
cd codex-claude-transfer && go build -o cct ./cmd/cct

Package manifests for Homebrew and Scoop live in packaging/.

cct doctor
cct export --project .
# copy project.codexbundle to the other machine
cct import ./project.codexbundle --dry-run
cct import ./project.codexbundle
# Optional: ask Codex to discover and verify changed threads immediately.
cct import ./project.codexbundle --reconcile

Use --tool claude to export Claude Code sessions. Use import --to claude or import --to codex to translate a session into the other agent. After importing, restart the agent so it re-scans the files. For a native Codex bundle, opt-in --reconcile can instead ask Codex's own app-server to read and verify the changed thread IDs immediately; failure leaves the imported rollout files intact and prints restart guidance plus an exact cct resume <thread-id> --run fallback only when the rollout ID is a valid UUID and the command can be rendered safely. The terminal wizard (cct ui) and browser app (cct app) expose the same opt-in native Codex reconciliation flow.

cct relocate rewrites the recorded working directory (cwd) in every matching session, so the sessions remain grouped with the project at its new path:

# The project was already copied or moved; NEW exists.
cct relocate /old/project /new/project --dry-run
cct relocate /old/project /new/project

# Move the project too; NEW must not exist yet.
cct relocate /old/project /new/project --move-project

# Claude Code: transcripts also move into the folder encoding the new path.
cct relocate /old/project /new/project --tool claude

The command preserves session backups and never modifies Codex's SQLite database or ~/.claude.json. Add --include-archived to relocate archived Codex sessions too. For Claude Code, each transcript is written under the new project folder first and its original removed only afterward, so a session id is never duplicated; cct undo reverses both halves. See the usage guide for rollback behavior and same-filesystem moves.

cct skill install writes a skill into your Claude Code home that teaches the agent one workflow: save this project's sessions into git when you stop, restore them after a clone on the other machine.

cct skill install                       # ~/.claude/skills/cct-session-sync/
cct skill print --plain >> ~/.codex/AGENTS.md   # the same for Codex

The recommended layout keeps chat history out of the code repo: one private session-store repo holds every project's bundles, and each project commits only a small reference file pointing at it.

cct config set repo-sync-repo git@github.com:you/cct-sessions.git
cct skill init     # writes .cct/sessions.json + .cct/README.md — commit them
cct skill show     # where the history lives and the exact commands

~/cct-sessions/projects/my-app/
  claude/claude-all.codexbundle      # every session for this project
  claude/groups/auth-refactor.codexbundle   # optional: one topic per file
  codex/codex-all.codexbundle

Without any agent, it is still just two commands — save with cct export -o <that path> and commit, restore with cct import <that path> --merge --map-cwd-here. Keeping the bundle in the project's own repo under .cct/ stays supported for private repos.

A bundle is readable by everyone with repo access, forever, so the skill asks once whether to commit it plainly (private repos only) or age-encrypted, and stores the answer in cct config. It never pushes or passes --allow-secrets on its own, and treats a reference file it did not write as untrusted. See the usage guide.

Codex's and Claude Code's on-disk formats are their own internals and can change at any time. This table records what each agent's support covers and the newest agent version it was last verified against:

If a newer agent version breaks something, please open an issue — with synthetic session data only, never a real bundle (see SECURITY.md).

• Usage guide: quickstart, desktop app, common workflows, LAN sync, and optional external tools.
• Command reference: commands and flags.
• Internals: how bundles work, safety model, limitations, and versioning.
• Safety notes: detailed privacy and write-safety model.
• Roadmap: shipped features, never-planned features, and project notes.

The daily Repository traffic metrics workflow stores a public snapshot at metrics/traffic.json on the separate metrics branch. It records GitHub's 14-day clone totals and merges the daily counts into an idempotent history. It also records the cumulative download count of installable release binaries and one daily snapshot of the displayed metrics. The current release total still covers every published release; download deltas and charts are complete only from the first stored daily snapshot. The cumulative clone count is complete only from tracked_since; GitHub does not expose older clone traffic. Daily unique-cloner counts are retained for auditability, but they are never summed into an all-time unique value.

The workflow needs one repository secret named CCT_TRAFFIC_TOKEN. Create it once as follows:

1. Open GitHub Settings → Developer settings → Personal access tokens → Fine-grained tokens and choose Generate new token.
2. Select the ahmojo resource owner, choose Only select repositories, and select only ahmojo/codex-claude-transfer.
3. Under Repository permissions, set Administration to Read-only. Do not grant write access or access to other repositories.
4. Generate the token and copy it immediately. GitHub shows it only once.
5. In ahmojo/codex-claude-transfer, open Settings → Secrets and variables → Actions → New repository secret.
6. Name the secret exactly CCT_TRAFFIC_TOKEN, paste the token, and save it.
7. Run Actions → Repository traffic metrics → Run workflow once. The normal GITHUB_TOKEN writes only the generated file to the metrics branch.

The token is sent only to GitHub's protected traffic endpoint. It is never written to the metrics file, commits, workflow logs, or a browser response. The generated file is publicly available at:

https://raw.githubusercontent.com/ahmojo/codex-claude-transfer/metrics/metrics/traffic.json

Clone traffic can include automated access from CI systems, scanners, bots, and other services. No attempt is made to estimate or subtract that traffic.

PRs welcome. Keep changes small, preserve the no-cloud / no-SQLite-writes principles, test with fake Codex homes only, and run the documented Go checks. See CONTRIBUTING.md.

Unofficial and not affiliated with or endorsed by OpenAI or Anthropic. It works against Codex's and Claude Code's local files based on their behavior at a point in time, which may change and break it. .codexbundle files can contain prompts, code, command output, paths, and secrets; treat them like sensitive local history and encrypt them over untrusted channels. Provided "as is", without warranty. Use at your own risk.