Data and portability commands

signet export / signet import

Export and import portable Signet bundles. This is the supported path for moving an agent between machines or backing up identity + memory state from the CLI.

signet export
signet export --json
signet export bundle
signet import ./signet-export-2026-03-22
signet import ./signet-export-2026-03-22.json --json --conflict merge

signet export (equivalent to signet export bundle) writes a portable bundle containing:

• identity files
• agent.yaml
• memories
• entities
• relations
• installed skills

signet import restores those files into $SIGNET_WORKSPACE/. Conflict handling for memories is controlled with --conflict:

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

---

signet export transcripts

Export stored session transcripts as JSONL (one conversation per line) for training and fine-tuning pipelines. Output is written to stdout by default so it can be piped, or to a file with --output.

# Export all transcripts as JSONL to stdout
signet export transcripts

# Write to a file
signet export transcripts --output ./training-data/conversations.jsonl

# Filter by harness and agent
signet export transcripts --harness hermes-agent --harness codex --agent ant

# Filter by creation date range
signet export transcripts --since 2026-06-01 --until 2026-07-01

# Resumable export
signet export transcripts --limit 5000 --offset 5000 --output ./part-2.jsonl

# Skip system and tool messages
signet export transcripts --messages-only

# Pretty-printed JSON array instead of JSONL
signet export transcripts --json

Each line has the shape:

{
  "id": "signet-db-<session-key>",
  "source": "signet",
  "harness": "hermes-agent",
  "agent_id": "ant",
  "session_key": "<session-key>",
  "project": "/path/to/project",
  "timestamp": "2026-07-01T12:00:00.000Z",
  "message_count": 12,
  "messages": [
    { "role": "user", "content": "..." },
    { "role": "assistant", "content": "..." }
  ]
}

The id is derived from the session key, so re-exporting the same data produces identical ids and downstream dedup by id stays stable. Transcript content is parsed as JSONL first (one {role, content} object per line); content without JSONL structure is parsed as role-prefixed text (User:/Assistant:/System:/tool_result: lines, with continuation lines accumulated into the previous message). Carriage returns split lines exactly like the training-data aggregator’s parser, so export output is a drop-in replacement for its SQLite reader.

Date boundaries: --since 2026-07-01 includes the whole of July 1; --until 2026-07-31 is treated as the end of July 31 (a date-only until compared raw would exclude the entire until day).

Options:

Flag	Default	Purpose
-o, --output <path>	stdout	Write to a file instead of stdout
--harness <name>	all	Filter by harness (repeatable)
--agent <name>	all	Filter by agent ID (repeatable)
--since <iso>	earliest	Only transcripts created at or after this ISO timestamp
--until <iso>	now	Only transcripts created at or before this ISO timestamp
--limit <n>	unlimited	Max conversations to export
--offset <n>	0	Skip N conversations (for resumable export)
--messages-only	false	Drop system and tool messages
--json	false	Output a JSON array instead of JSONL

---

signet migrate-schema

Migrate an existing memory database to Signet’s unified schema. Useful when upgrading from an older version or copying $SIGNET_WORKSPACE/ between machines.

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

Supported source schemas:

Schema	Source
python	Original Python memory system
cli-v1	Early Signet CLI (v0.1.x)
core	Current unified schema (no migration needed)

Migration is idempotent - safe to run multiple times. All existing memories are preserved. The daemon is stopped and restarted automatically during the process.

Output:

- Checking database schema...
  Migrating from python schema...
  ✓ Migrated 261 memories from python to core

  Migration complete!

---

signet migrate-vectors

Migrate existing BLOB-format embeddings to the sqlite-vec format. Run this once after upgrading from a version that stored vectors as raw BLOBs.

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

Options:

Option	Description
--keep-blobs	Keep the old BLOB column after migration (safer rollback)
--remove-zvec	Delete vectors.zvec file after successful migration
--dry-run	Show what would be migrated without making changes

---

signet embed

Manage memory embeddings. Requires the daemon to be running.

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

Subcommands:

Command	Description
signet embed backfill	Generate missing vectors; use --model-mismatch for a model/dimension migration or --all to force active memories
signet embed gaps	Show count of memories missing embeddings

signet embed backfill options:

Option	Description
--batch-size <n>	Memories per batch (default: 50)
--dry-run	Preview without calling the embedding provider
--model-mismatch	Re-embed vectors whose stored model or dimensions differ from the configured target
--all	Re-embed all active memories in bounded resumable batches

Migration dry runs report source model/dimension labels (historical provider identity is not recorded), the configured target, estimated batches, and whether dimensions require a vector-index rebuild.

After backfill completes, coverage is printed:

  Coverage: 100.0% (1200/1200 embedded)

---