Skip to main content

Privacy & Data Handling

Your full trace files and canonical execution facts stay on your machine. The CLI processes the full dataset locally. The service receives representative samples, filtered by your redaction settings, plus a schema/profile, registration metadata, and bounded evidence computed against the local data. Optional workflow files are sent in full.

This page describes the CLI analysis flow. Browser uploads send trace data to the service and have a different data path. See Scail's Privacy Policy and Security information for service policies. Scail uses supplied data to provide the analysis and does not use it to train general-purpose AI models.

Local data and outbound data​

DataCLI behavior
Full trace file or connector exportStays on your machine.
Canonical execution facts / ParquetStays on your machine; queried locally for evidence.
Trace profile and representative samplesSent for normalization after the configured field-name redaction. Samples can contain trace content.
Registration metadataHashes, counts, workflow configuration, and other submission metadata are sent to register the analysis.
Bounded evidenceAllowlisted columns and bounded query results are sent during investigation.
Workflow attachmentsFull text and relative filenames are sent to the service and its models; bounded excerpts can be inspected during investigation.

The CLI requires an internet connection. The service uses models for normalization and analysis. Keeping full trace files local does not mean no trace content leaves your machine.

Redact trace samples​

Choose field names that match your export:

jejoo analyze traces.jsonl \
--remove-key prompt \
--remove-key response \
--hash-key customer_id \
--hash-salt-env JEJOO_HASH_SALT \
--redaction-policy-id company-policy-v1 \
--block-column error_type \
--api-base-url https://app.scaillabs.ai \
--output-dir .jejoo/redacted-analysis

Set JEJOO_HASH_SALT privately before running the example. --remove-key recursively removes matching field names from normalization samples; --hash-key replaces matching values with SHA-256 hashes. A key cannot be both removed and hashed. --redaction-policy-id labels the policy; the label alone changes no fields.

These controls do not automatically detect secrets embedded in arbitrary text. Sample redaction does not redact workflow attachments or replace the separate evidence column policy. Review workflow files before supplying them.

Evidence uses a built-in column allowlist. Repeat --block-column to exclude columns, or --allow-column to narrow the allowlist. You cannot enable a column outside the built-in allowlist. Evidence row, byte, and request limits are documented in the CLI Reference.

For staged inspection before a network request, use jejoo local prepare --help to write a normalization request and private local state. Running analyze performs submission as part of the command.

Credentials, results, and retention​

The CLI stores credentials under ~/.config/jejoo by default with owner-only permissions. Keep that directory private. Valid legacy credentials from ~/.config/optimaize are copied when first used and retained there for rollback. An explicit configuration directory does not migrate credentials from another directory.

Anonymous analysis jobs and their job credentials expire after 24 hours. Background normalization metadata/results expire after 24 hours. Authenticated analysis data defaults to 30 days, subject to deployment configuration. Expired service data is processed by a scheduled purge; expiry is not a promise of deletion at the exact second. Service retention does not delete files on your computer.

Keep local exports, output directories, and credential backups out of source control and shared folders. Delete local results yourself when you no longer need them.