Skip to content

Export your sessions and remove paper

Take your organization's complete session history with you, keep the skills you published, remove paper from your machines, and keep capturing agent sessions with open source tapes.

Everything you need to take your data with you and leave a machine clean: export your organization’s complete session history, keep the skills you published, remove paper from your machines, and, if you want to keep capturing agent sessions, pick up tapes, the open source engine behind paper.

Export your organization’s sessions

Two ways to do it. Both produce the same NDJSON file, one line per session, described in What the export gives you. Export reaches your complete history back to your first session, on every plan.

From the console. Open Sessions, choose Export sessions, then Download everything. The download is named sessions-all-<date>.jsonl and covers the current gateway. To export one session, use Export on its detail page.

From the command line (recommended for large organizations and for scripting). Update paperctl, make sure you are signed in, then export:

paperctl upgrade
paperctl status          # auth must be healthy; if it says expired, run paperctl login
paperctl sessions export --all --out everything.jsonl

--all returns everything the organization has recorded, back to its first session. It streams to the file as it goes, so a large organization’s export takes a few minutes and the file grows while it runs. With expired auth the command fails fast with no active org (run paperctl login …); sign in and rerun.

A whole-history export is large: one organization’s 5,700 sessions ran past 4 GB, streaming at roughly 100 MB a minute. Check free disk space first (df -h .), and if space is tight, export in time windows, or use --detail traces for a far smaller file without span content.

Smaller files are available too:

# A time window (RFC 3339, UTC)
paperctl sessions export --since 2026-06-01T00:00:00Z --until 2026-07-01T00:00:00Z --out june.jsonl

# One session
paperctl sessions export <session-id> --out session.jsonl

# Headers only: every session and trace, without the spans — a much smaller file
paperctl sessions export --all --detail traces --out everything-traces.jsonl

An export covers one gateway. If your organization has more than one (paperctl tapes gateway list names them, and paperctl status shows the active one), export each gateway separately:

paperctl tapes gateway list
paperctl --gateway-id default sessions export --all --out default.jsonl
paperctl --gateway-id <other-gateway> sessions export --all --out other.jsonl

An empty file for a window with no sessions is the correct answer, not a failure. Export session data has the full reference.

Save the skills you want to keep

A synced skill is a plain SKILL.md on disk and keeps working without paper, so sync each one you care about. Sync takes the skill’s slug, which the JSON listing shows:

paperctl skill list --json | jq -r '.items[].slug'
paperctl skill sync <slug> --claude   # ~/.claude/skills/<slug>/SKILL.md
paperctl skill sync <slug>            # ~/.agents/skills/<slug>/SKILL.md

Expected: Synced <slug> to …/SKILL.md. A skill whose latest revision is private is refused with HTTP 409 … (revision_private); open it on the console’s Skills page and save its content by hand. Install a Skill covers per-project targets and other agents.

Remove paper from your machines

paperctl menu uninstall                 # macOS menu bar app, if you installed it
paperctl plugin uninstall codex-app     # if Codex or the ChatGPT app ever ran through paper
paperctl uninstall

paperctl uninstall removes the daemon, your credentials, the installer’s shell block, and the binaries. A few things live outside paper’s own files and need a hand: the Pi extension, a shell routing line you added yourself, Conductor and Zed environment settings, and the skills paperctl skills install wrote. Uninstall paperctl walks through every one of them, with a checklist of paths to verify and a manual path for a machine where paperctl is already gone.

Keep capturing with tapes

tapes is the open source engine that recorded your paper sessions. Run it yourself and keep the same capture workflow:

curl -fsSL https://download.tapes.dev/install | bash
curl -sSfL https://download.tapes.dev/tapesctl/install | bash
tapes local up && tapes serve
tapesctl start claude --ingest-url http://localhost:8082

tapesctl start claude, tapesctl start codex, and tapesctl start pi work the way paperctl start did, and tapesctl also captures Cursor and the ChatGPT app. The installation guide covers the local PostgreSQL and Ollama setup, and integrations covers each agent. If you used Pi through paper, tapesctl reuses the same extension file, so install tapesctl before deleting it.

What the export gives you

The console and the CLI produce the same file: NDJSON, one line per session. Each line is one JSON object with these keys:

  • schema — the record format version (2026-06-15 today).
  • session — the session record: id, harness, working directory, timestamps, and metadata.
  • traces — the session’s traces in order. Each entry holds a trace record (one turn) and, at full detail, its spans (the model calls, tool calls, and their content).
  • links — at full detail only, the links between spans across traces.

What varies is the detail level:

Source Detail What each line holds
Console, Export on a session, Download or Download everything on the list full session + traces, each trace with its spans and content
paperctl sessions export … (default, --detail spans) full the same
paperctl sessions export … --detail traces headers session + traces, each trace without its spans — the shape of the history without its content, in a far smaller file

The console has no detail switch; it always exports full detail. Scripts that read traces[].spans should run against a full-detail file. To see what you have:

wc -l everything.jsonl                       # one line per session
head -1 everything.jsonl | jq 'keys'         # ["links","schema","session","traces"] at full detail
head -1 everything.jsonl | jq '.traces[0] | keys'   # ["spans","trace"] — or ["trace"] with --detail traces

Frequently asked questions

Can I export my whole session history?+
Yes. Export reaches your organization's complete history back to its first session, on every plan. Use Download everything in the console, or paperctl sessions export --all from the terminal.
What format is an export?+
NDJSON, one line per session, from both the console and the CLI. Each line holds the session record and its traces; at full detail (the console, and the CLI default) every trace carries its spans, while the CLI's --detail traces leaves the spans out. See What the export gives you.
Do skills I synced keep working after I remove paper?+
Yes. A synced skill is a plain SKILL.md on disk and does not call paperctl. Only the recall skills written by paperctl skills install depend on the CLI.
How do I keep capturing sessions without paper?+
Run tapes, the open source engine behind paper. tapesctl start claude, tapesctl start codex, and tapesctl start pi work the way paperctl start did, against a tapes server you run yourself.
Copied to clipboard