Save Memory - Manual Session Summary Storage

Save the current session context to the AI Memory system's discussions collection. This creates a type=session entry that can be retrieved by SessionStart on future session resume or by /aim-search.

When to Use

• Before ending a session to preserve important context
• After making significant decisions you want remembered
• When you want to bookmark a particular conversation state
• As a manual complement to the automatic PreCompact session save

Usage

The save-memory command runs the manual_save_memory.py script using the project's configured AI Memory installation.

# Save with a description (recommended)
"$AI_MEMORY_INSTALL_DIR/.venv/bin/python" "$AI_MEMORY_INSTALL_DIR/.claude/hooks/scripts/manual_save_memory.py" \
"Completed authentication refactor, decided on JWT approach"

# Save without description
"$AI_MEMORY_INSTALL_DIR/.venv/bin/python" "$AI_MEMORY_INSTALL_DIR/.claude/hooks/scripts/manual_save_memory.py"

What Gets Stored

The script stores to the discussions collection with:

• type: session (default); agent_memory or agent_insight (when --type is used)
• source_hook: ManualSave
• group_id: Auto-detected from current project directory
• session_id: Current Claude session ID (from CLAUDE_SESSION_ID env var)
• content: Structured summary with timestamp and user description
• embedding: 768-dim Jina v2 vector for semantic retrieval
• content_hash: SHA-256 for deduplication

Agent Memory Support

When --type agent_memory or --type agent_insight is used, the memory is stored via store_agent_memory() to the Parzival namespace with agent_id=parzival.

This requires Parzival to be enabled (parzival_enabled=true in config). If Parzival is not enabled, the command returns an error.

Activation

# Default behavior (unchanged)
/aim-save "Completed authentication refactor"

# Save as agent memory (general project knowledge)
/aim-save "The decay formula uses 0.7/0.3 weighting" --type agent_memory

# Save as agent insight (key learning or pattern)
/aim-save "PyYAML not in test deps caused CI failures" --type agent_insight

Environment Variables (Auto-Configured)

These are set automatically in settings.json by the installer:

• AI_MEMORY_INSTALL_DIR — Path to ~/.ai-memory installation
• AI_MEMORY_PROJECT_ID — Current project name
• QDRANT_HOST / QDRANT_PORT — Qdrant connection (localhost:26350)
• QDRANT_API_KEY — Qdrant authentication
• EMBEDDING_HOST / EMBEDDING_PORT — Embedding service (127.0.0.1:28080)

Error Handling

• If Qdrant is unavailable: queues to ~/.ai-memory/queue/ for background processing
• If embedding fails: stores with zero vector (backfilled later)
• Exit code 0 on success OR successful queue fallback
• Exit code 1 only on hard errors (missing installation)

Prerequisites

• AI Memory services running (docker compose up -d from ~/.ai-memory/docker/)
• Installation directory exists at $AI_MEMORY_INSTALL_DIR
• Python venv at $AI_MEMORY_INSTALL_DIR/.venv/

Activation Examples

# Save decision context
/aim-save "Decided to use Qdrant for vector storage over Pinecone due to self-hosting requirement"

# Save progress checkpoint
/aim-save "Completed 3 of 5 API endpoints, auth middleware working"

# Quick save without description
/aim-save

# Save as agent memory (general project knowledge)
/aim-save "The decay formula uses 0.7/0.3 weighting" --type agent_memory

# Save as agent insight (key learning or pattern)
/aim-save "PyYAML not in test deps caused CI failures" --type agent_insight

Technical Details

• Script: .claude/hooks/scripts/manual_save_memory.py
• Collection: discussions (COLLECTION_DISCUSSIONS)
• Type: session (same as PreCompact auto-saves)
• Embedding: jina-embeddings-v2-base-en (768 dimensions)
• Fallback: File queue at ~/.ai-memory/queue/ via queue_operation()
• Logging: Activity logged via log_manual_save() to ~/.ai-memory/logs/activity.log