`mayros ltm`

Manage the LanceDB-backed long-term vector memory. List stored memories, perform semantic searches, and view memory statistics.

Memories are automatically captured from conversations (preferences, decisions, entities, facts) and recalled before each agent session when auto-recall is enabled.

Subcommands

`ltm list`

List the total number of memories stored in the LanceDB database.

`ltm search <query>`

Perform a semantic vector search across stored memories. The query is embedded using the configured OpenAI embedding model and compared against all stored memory vectors.

Options:

Flag	Description
`--limit <n>`	Maximum results to return (default: `5`)

Results are returned as JSON with: id, text, category, importance, and score (similarity 0-1).

`ltm stats`

Show memory statistics including the total count of stored memories.

Memory categories

Memories are automatically categorized:

Category	Examples
`preference`	"I prefer dark mode", "I like TypeScript"
`decision`	"We decided to use PostgreSQL"
`entity`	Email addresses, phone numbers, named entities
`fact`	General factual statements
`other`	Uncategorized memories

Configuration

The memory plugin is configured in mayros.toml under the memory-lancedb section:

Key	Description
`dbPath`	LanceDB database path (resolved relative to project)
`embedding.apiKey`	OpenAI API key for embeddings
`embedding.model`	Embedding model (default: `text-embedding-3-small`)
`autoRecall`	Inject relevant memories before agent start
`autoCapture`	Automatically capture important info from conversations
`captureMaxChars`	Maximum text length for auto-capture

Examples

bash
mayros ltm list
mayros ltm search "database preferences"
mayros ltm search "authentication decisions" --limit 10
mayros ltm stats

Memory — memory architecture overview
Agent Memory — per-agent persistent memory