Skip to main content

​
Overview

The memory reindex command regenerates embeddings for all memories using the current embedding provider configured in config.yaml. This is necessary after changing embedding settings or when adding embeddings to an existing vault.

​
Syntax

memory reindex

​
When to Use

​
After changing embedding provider

When you switch providers: 
# Edit config.yaml to change provider from ollama to openai
nano ~/.memory/config.yaml

# Rebuild all embeddings
memory reindex

​
After changing embedding model

When you switch models within the same provider: 
# Edit config.yaml to change from nomic-embed-text to mxbai-embed-large
nano ~/.memory/config.yaml

# Reindex with new model
memory reindex

​
Adding embeddings to existing vault

When you initially used keyword-only search and now want semantic search: 
# Configure embeddings in config.yaml
memory config init
nano ~/.memory/config.yaml

# Generate embeddings for all existing memories
memory reindex

​
After embedding provider upgrade

When your provider releases a new model version: 
# Update model in config.yaml
memory reindex

​
Examples

​
Basic reindex

memory reindex
Output: 
Reindexing 47 memories with ollama/nomic-embed-text...
  47/47
Re-indexed 47 memories with nomic-embed-text (768 dims)

​
When no memories exist

memory reindex
Output: 
No memories to reindex.

​
With progress updates

For large vaults, you’ll see progress: 
memory reindex
Output: 
Reindexing 523 memories with openai/text-embedding-3-small...
  100/523
  200/523
  300/523
  400/523
  500/523
  523/523
Re-indexed 523 memories with text-embedding-3-small (1536 dims)

​
How It Works

  1. Count memories: Queries the database for total memory count
  2. Load config: Reads current embedding provider and model from config.yaml
  3. Generate embeddings: For each memory:
    • Extracts text (title + what + why + impact + details)
    • Sends to embedding provider API
    • Stores the vector in the database
  4. Show progress: Updates the counter for each memory processed
  5. Report results: Shows total count, model name, and vector dimensions

​
Performance

​
Speed

  • Ollama (local): 10-50 memories/second (depends on hardware)
  • OpenAI: 5-20 memories/second (depends on rate limits)

​
Time estimates

  • 50 memories: ~5-15 seconds
  • 200 memories: ~20-60 seconds
  • 1000 memories: ~2-5 minutes

​
Rate limiting

If you hit rate limits: 
memory reindex
# Wait for error, then retry:
memory reindex  # It will continue from where it left off

​
Supported Providers

​
Ollama (local)

embedding:
  provider: ollama
  model: nomic-embed-text  # or mxbai-embed-large
  base_url: http://localhost:11434
Models:
  • nomic-embed-text: 768 dimensions, fast, good quality
  • mxbai-embed-large: 1024 dimensions, slower, better quality

​
OpenAI

embedding:
  provider: openai
  model: text-embedding-3-small
  base_url: https://api.openai.com/v1
  api_key: sk-...
Models:
  • text-embedding-3-small: 1536 dimensions, fast, cheaper
  • text-embedding-3-large: 3072 dimensions, slower, more expensive
  • text-embedding-ada-002: 1536 dimensions (legacy)

​
Error Handling

​
Provider not configured

memory reindex
Output: 
Error: No embedding provider configured in config.yaml
Fix: 
memory config init
nano ~/.memory/config.yaml
# Configure embedding section
memory reindex

​
Provider unavailable

memory reindex
Output: 
Error: Failed to connect to ollama at http://localhost:11434
Fix: 
# Start Ollama
ollama serve

# In another terminal:
memory reindex

​
Invalid API key

memory reindex
Output: 
Error: Authentication failed with OpenAI API
Fix: 
nano ~/.memory/config.yaml
# Update api_key
memory reindex

​
What Gets Updated

​
Database changes

  • embeddings table: All vectors are replaced
  • model metadata: Updated with new model name and dimensions
  • memories table: Unchanged (content is preserved)

​
Files unchanged

  • Memory markdown files in vault/ are NOT modified
  • Only the database vectors are regenerated

​
Use Cases

# You've been using EchoVault with FTS-only
# Now you want semantic search

# 1. Configure embeddings
memory config init
nano ~/.memory/config.yaml

# 2. Generate vectors for existing memories
memory reindex

# 3. Test semantic search
memory search "why did we choose postgres"
memory context --semantic --project

​
Switch from Ollama to OpenAI

# Edit config
nano ~/.memory/config.yaml
# Change provider: ollama → openai
# Add api_key

# Rebuild with OpenAI
memory reindex

# Verify
memory config | grep embedding

​
Upgrade to better model

# Edit config
nano ~/.memory/config.yaml
# Change model: nomic-embed-text → mxbai-embed-large

# Rebuild
memory reindex

​
Fix corrupted embeddings

If embeddings are corrupted or inconsistent: 
memory reindex
This regenerates all vectors from scratch.

​
Verification

​
Check embedding count

sqlite3 ~/.memory/memory.db "SELECT COUNT(*) FROM embeddings;"

​
Check model metadata

memory config

memory search "natural language query about your project"
memory context --semantic --project
After reindexing, use memory search with natural language queries to verify semantic search is working. Try queries like “why did we…” or “how do we handle…”
Reindexing is safe and idempotent. You can run it multiple times without risk. It only updates the vector database, never modifies your memory files.

Build docs developers (and LLMs) love

Get started for freeTalk to us