Export & import

Palimem provides operator tools for exporting memory (for review or compliance) and importing from Markdown files (for seeding or migration).

---

Export for review

The memory_review tool (MCP) and ai-memory review export operator CLI generate a human-readable review report of current memory state:

Operator CLI (recommended)

node app/scripts/ai-memory.js review export \
  --data-dir .ai-memory/data \
  --output .ai-memory/review.md

This writes a Markdown file with all current facts, beliefs, and preferences grouped by scope and namespace.

MCP tool

{
  "tool": "memory_review",
  "scope": "repository",
  "namespace": "my-project",
  "action": "export"
}

Review report format

# Memory review — my-project (repository scope)
Generated: 2026-06-25T12:00:00Z

## auth
- provider: OAuth2 (fact, seq 42)
- expiry: 3600s (fact, seq 43)

## deploy
- region: us-east-1 (fact, seq 51)
- service: ECS Fargate (fact, seq 52)

---

Audit export (compliance)

memory_audit_export produces a compliance-ready structured export of the full WAL, with provenance metadata:

{
  "tool": "memory_audit_export",
  "scope": "repository",
  "namespace": "my-project",
  "from": "2026-01-01T00:00:00Z",
  "to": "2026-06-30T23:59:59Z"
}

The export includes:

• All WAL events (remember, supersede, retract)
• Provenance: source, tool, actor, request_id
• Timestamps and sequence numbers
• Legal hold flags

Use audit export for:

• Compliance reviews (“what did the agent know and when?”)
• Debugging unexpected recall behavior
• Handoff to a new environment

---

Import from Markdown

Seed memory from Markdown files using import_markdown.py. This is the standard path for:

• Migrating from file-based memory systems
• Pre-loading user profile and project conventions
• Seeding from MEMORY.md or USER.md files

python3 app/import_markdown.py \
  --data-dir .ai-memory/data \
  examples/markdown/USER.md.sample \
  examples/markdown/MEMORY.md.sample

Sample files

The repository ships sample Markdown files:

File	Purpose
examples/markdown/USER.md.sample	User preferences and profile facts
examples/markdown/MEMORY.md.sample	Project memory conventions

Copy and edit these samples for your project, then import.

---

Consolidation

Consolidation deduplicates redundant units and summarizes low-salience noise clusters. It produces proposals that must be reviewed before becoming current facts.

1. Run dry-run to see proposals

   node app/scripts/ai-memory.js consolidate \
     --data-dir .ai-memory/data \
     --dry-run

2. Export review report

   node app/scripts/ai-memory.js review export \
     --data-dir .ai-memory/data \
     --output .ai-memory/review.md

3. Accept proposals (MCP tool or operator CLI)

   node app/scripts/ai-memory.js consolidate \
     --data-dir .ai-memory/data \
     --export-review .ai-memory/review.md

Consolidation never removes facts under legal hold or current semantic units. It is reversible — the WAL is unchanged.

---

Cron-friendly consolidation

Run consolidation on a schedule from a shell script:

# Run every night
bash examples/claude-code/demo/run-consolidation-cron.sh

Or call directly:

node app/scripts/ai-memory.js consolidate \
  --data-dir .ai-memory/data \
  --export-review .ai-memory/review.md

---

Import from review export

After editing a review Markdown file, re-import accepted changes:

python3 app/import_markdown.py \
  --data-dir .ai-memory/data \
  .ai-memory/review.md

---

Point-in-time recall

Use as_of on memory_get and memory_search to query historical state:

{
  "tool": "memory_get",
  "scope": "repository",
  "namespace": "my-project",
  "topic": "auth",
  "field": "provider",
  "memory_type": "fact",
  "as_of": "2026-01-15T09:00:00Z"
}

---

Next steps

• Governance — retention, PII policy, and legal hold
• Supersession & audit — how the immutable WAL enables point-in-time recall
• MCP tools reference — full arguments for memory_review, memory_consolidate, memory_audit_export