embedding-ops

name	embedding-ops
description	Configure, test, benchmark, and manage embedding providers for semantic search and vector operations

Embedding Operations

When to Use

• Configuring embedding providers (OpenAI, Cohere, Ollama, local)
• Testing embedding connectivity and quality
• Benchmarking provider performance
• Rebuilding embedding indices
• Debugging semantic search issues

CLI Commands

Command	Purpose
do-memory-cli embedding config	Show current embedding configuration
do-memory-cli embedding test	Test embedding provider connectivity
do-memory-cli embedding list-providers	List available providers
do-memory-cli embedding benchmark	Benchmark provider latency/throughput
do-memory-cli embedding rebuild	Rebuild embedding index from episodes
do-memory-cli embedding enable	Enable embeddings (session only)
do-memory-cli embedding disable	Disable embeddings (session only)

MCP Tools

Tool	Parameters	Purpose
configure_embeddings	provider, model, dimension, api_key_env	Configure embedding provider
query_semantic_memory	query, limit, threshold	Semantic similarity search
test_embeddings	text	Test embedding generation
generate_embedding	text, provider	Generate embedding vector
search_by_embedding	embedding, limit	Search by raw vector
embedding_provider_status	-	Check provider health

Configuration

Set in config file or environment:

[embeddings]
enabled = true
provider = "openai"  # openai | cohere | ollama | local
model = "text-embedding-3-small"
dimension = 1536
api_key_env = "OPENAI_API_KEY"

Provider Setup

OpenAI

export OPENAI_API_KEY="sk-..."
do-memory-cli embedding config
do-memory-cli embedding test

Ollama (local)

ollama pull nomic-embed-text
# Config: provider = "ollama", model = "nomic-embed-text", dimension = 768

Troubleshooting

1. Test fails: Check API key env var is set, provider is reachable
2. Slow search: Run benchmark to identify latency; consider local provider
3. Poor results: Check dimension matches model; rebuild index after provider change

Known Limitations

• enable/disable commands affect current session only (no persistence)
• --semantic-search flag on episode list is declared but not yet wired
• HNSW index save() is a no-op when hnsw feature is enabled