Memory

Memory

Milady supports pluggable memory backends. Configuration lives under the memory key in ~/.milady/milady.json.

The schema is defined in src/config/types.milady.ts (MemoryConfig, MemoryQmdConfig).

---

Backends

Backend	Value	Description
Built-in	"builtin"	elizaOS default memory (default)
QMD	"qmd"	Local file-based semantic memory with embedding search

{
  "memory": {
    "backend": "builtin"
  }
}

---

Citations mode

{
  "memory": {
    "citations": "auto"
  }
}

Valid values: "auto", "on", "off".

---

QMD backend

QMD provides local file-based semantic memory. It uses embedding search to retrieve relevant memories and injects them into context.

{
  "memory": {
    "backend": "qmd",
    "qmd": {
      "command": "qmd",
      "includeDefaultMemory": true,
      "paths": [
        { "path": "~/notes", "name": "notes", "pattern": "**/*.md" }
      ],
      "sessions": {
        "enabled": true,
        "exportDir": "~/.milady/session-exports",
        "retentionDays": 30
      },
      "update": {
        "interval": "1h",
        "debounceMs": 2000,
        "onBoot": true,
        "embedInterval": "6h"
      },
      "limits": {
        "maxResults": 20,
        "maxSnippetChars": 500,
        "maxInjectedChars": 4000,
        "timeoutMs": 5000
      }
    }
  }
}

QMD fields

Field	Description
command	QMD CLI command
includeDefaultMemory	Include elizaOS default memory alongside QMD results
paths	Index paths: path, optional name, optional pattern
sessions.enabled	Index session transcripts
sessions.exportDir	Where to export session transcripts
sessions.retentionDays	How long to keep exported sessions
update.interval	How often to re-index
update.debounceMs	Debounce for file watcher
update.onBoot	Re-index on startup
limits.maxResults	Max memory snippets to inject
limits.maxSnippetChars	Max chars per snippet
limits.maxInjectedChars	Total injected memory cap
limits.timeoutMs	Memory query timeout

---

Vector memory search (per-agent)

Per-agent memory search config lives under agents.defaults.memorySearch or per-agent under agents.list[].memorySearch. Schema: MemorySearchConfig in src/config/types.tools.ts.

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "enabled": true,
        "sources": ["memory", "sessions"],
        "provider": "local",
        "store": {
          "driver": "sqlite",
          "vector": { "enabled": true }
        },
        "query": {
          "maxResults": 10,
          "minScore": 0.7,
          "hybrid": {
            "enabled": true,
            "vectorWeight": 0.7,
            "textWeight": 0.3
          }
        },
        "sync": {
          "onSessionStart": true,
          "watch": true,
          "intervalMinutes": 60
        }
      }
    }
  }
}

Embedding providers

Value	Description
"local"	Local GGUF model via node-llama-cpp
"openai"	OpenAI embeddings API
"gemini"	Google Gemini embeddings API

Local embedding model config lives under the top-level embedding key:

{
  "embedding": {
    "model": "nomic-embed-text-v1.5.Q5_K_M.gguf",
    "dimensions": 768,
    "contextSize": 2048,
    "gpuLayers": "auto",
    "idleTimeoutMinutes": 30
  }
}