2.13.0-beta.20). Features may change before the next stable release.Switch to stable →ak analytics
ak analytics delete
Remove AgentKit-owned derived analytics files while retaining a disabled intent marker and all authoritative source data.
Use ak analytics delete to remove the disposable local analytics database.
It does not delete Claude Code sessions, Codex sessions, or the AgentKit
activity log.
Usage
ak analytics deleteThe command accepts no positional arguments and has no command-specific flags.
Preview and apply
ak analytics delete --json --no-interactive
ak analytics delete --yes --json --no-interactivePreview reads lifecycle and status only; it does not scan source contents or
write. Apply removes AgentKit-owned files under
~/.agentkit/analytics/analytics.db, including SQLite sidecars, and purges the
analytics recovery directory. It retains only
~/.agentkit/analytics-state.json with deleted intent so indexed reads and
automatic reconciliation stay disabled. AGENTKIT_HOME changes the base.
Delete is valid from any lifecycle state, including when no database exists, so it is safe to repeat.
Ownership and recovery boundaries
The operation refuses unsafe paths, a newer-schema database, foreign database content, or unrecognized non-database bytes. It can remove an owned corrupt database or an interrupted empty enable artifact when ownership can be established without trusting the damaged schema.
For a readable index, delete makes a consistent temporary rollback snapshot. For an owned unreadable index, it copies bounded sidecars. Those files exist only to restore a failed partial removal and are purged after success; they are not exposed as a user-restorable backup.
Successful deletion is therefore irreversible for the derived index. Because
sources remain authoritative, you can explicitly recreate it with
ak analytics enable --yes or ak analytics rebuild --yes.
Confirmation behavior
When prompting is enabled, only y or yes applies. Any other answer or EOF
prints the preview, changes nothing, and currently exits 0.
With --json or --no-interactive, no prompt is read. Without --yes, the
preview exits 3. There is no dry-run or force flag.
Output and privacy
Human output goes to stdout and the prompt to stderr. JSON uses one
kind=analytics.plan envelope with data.plan, data.applied, and
data.status; delete includes no source scan.
Use data.plan.status for a preview and the post-operation data.status only
when applied=true. The response and operation expose no source paths or raw
transcript content and perform no network, provider, auth, or entitlement work.
Exit behavior
| Exit | Meaning |
|---|---|
0 | Delete applied, was already effectively deleted, or an enabled prompt was declined/ended at EOF. |
1 | Planning, ownership/safety validation, contention, removal, rollback, or output failed; positional arguments also use 1. |
2 | A flag was invalid. |
3 | A noninteractive preview completed without --yes. |
Lifecycle/service failures in JSON mode emit a versioned error envelope on stderr. Flag-parse, argument-validation, and final output-render failures can fall outside that path.
On macOS, the CLI-wide first-launch guard can also write
~/.agentkit/.first-run-darwin and print a one-time hint to stderr before this
command runs.