Data and portability commands

Export and import a portable bundle

signet export
signet export --json
signet export bundle
signet import ./signet-export
signet import ./signet-export.json --json --conflict merge

signet export is equivalent to signet export bundle. Bundles carry the workspace identity files, agent.yaml, memory and ontology data, and installed skills. Import restores a bundle into the selected workspace.

Choose a conflict policy deliberately:

• skip: keep existing memories and skip duplicates.
• overwrite: replace matching records.
• merge: merge compatible records when supported.

Treat an export as sensitive: it can include private identity and memory data. Encrypt it at rest and do not upload it to an untrusted service.

Export transcripts

signet export transcripts
signet export transcripts --output ./conversations.jsonl
signet export transcripts --harness hermes-agent --agent ant
signet export transcripts --since 2026-06-01 --until 2026-07-01
signet export transcripts --limit 5000 --offset 5000 --output ./part-2.jsonl
signet export transcripts --messages-only
signet export transcripts --json

Transcript export writes JSONL to standard output by default, or to --output. Export requires the daemon and remains scoped to its resolved agent. Use repeated --harness filters and --agent to assert that scope, --limit and --offset for resumable exports, and --messages-only to omit system and tool messages. --json streams a JSON array instead of JSONL. An output file must not already exist.

Transcripts can contain credentials, personal data, and private project context. Review access controls before using an export for training or analysis.

Importing exported transcripts

signet sources import <files...> --kind transcripts --schema signet --agent <id> streams the current signet export transcripts JSONL shape into a durable daemon job. signet sources imports list|status|pause|resume|retry|cancel reads or controls that job. The daemon records per-line hashes, byte checkpoints, statuses, rejection codes, and reconciliation counters. Raw transcript bodies are retained separately as bounded database chunks.

The adapter is signet-export version 1. It preserves exact whitespace, multiline messages, user/assistant/system/tool/unknown roles, projects, historical timestamps, and source provenance. --agent is the target scope; an embedded agent_id is not an authorization or routing override. Same identity plus same content is duplicate; same identity plus changed content is conversation_identity_conflict. Blank lines are ignored.

Transcript imports and imported-source deletion support Windows, Linux, and macOS. The single database owner retains the raw bytes and resumes checksummed uploads. See the import API for upload limits, disk-space admission, and migration of older filesystem imports.

Resume an interrupted upload with the job and file IDs from status:

signet sources imports upload <job-id> <file-id> ./conversations.jsonl --agent <id>

The client verifies the local prefix against the durable checksum chain before continuing. Finalization makes one streaming pass over the stored file for its standard SHA-256 identity. Parsing starts only after evidence is sealed.

Limits are one active job/file, 25 records per DB batch, 8 MiB per canonical batch, 16 MiB per record, 4 MiB per message, and 50,000 messages. States are staging, inventorying, queued, running, paused, completed, completed_with_rejections, and cancelled. Restart recovers leases and byte offsets. The completed-job reconciliation invariant is `total = imported + duplicate

• rejected + pendingwithpending = 0`. Exact corpus replay produces duplicates, not new evidence. Source removal purges retained bytes and source rows, while bounded audit fingerprints/tombstones remain; derived knowledge is handled as unsupported/stale by the normal Dreaming review path.

Schema and vector migrations

signet migrate-schema
signet migrate-schema --path /custom/path

signet migrate-vectors --dry-run
signet migrate-vectors --keep-blobs
signet migrate-vectors --remove-zvec

Run migrations against a backup and make sure the intended daemon/workspace is selected. migrate-schema is designed to be idempotent for supported legacy schemas. migrate-vectors --dry-run previews a vector migration before it changes data.

Audit embedding coverage

signet embed audit
signet embed audit --json
signet embed backfill --dry-run
signet embed backfill --batch-size 100
signet embed backfill --model-mismatch --dry-run
signet embed backfill --all --dry-run

signet embed audit is the supported coverage command. It reports total, embedded, missing, and coverage values, and may include staging information. signet embed gaps is retired and is not a valid CLI subcommand.

Use signet embed backfill to generate missing embeddings. --model-mismatch selects vectors with a different stored model or dimension. --all re-embeds every active memory and therefore requires either --dry-run or --model-mismatch as an explicit migration confirmation.

Embedding commands require a reachable daemon. See Runtime operations if it is not running.