Environment Variables

This page is a quick-reference table of all AutoMem environment variables, grouped by category. For full documentation of each variable including validation rules, behavior details, and configuration examples, see Configuration Reference.

Configuration Loading

Variables are loaded from multiple sources in priority order:

Process environment — export VAR=value or variables injected by the platform (Railway, Docker)
Project .env — File in repository root
User config — ~/.config/automem/.env (personal overrides, never committed)

Variables set earlier in this list take precedence over later sources.

Server Variables

Core Database

Variable	Required	Default	Description
`FALKORDB_HOST`	Yes	`localhost`	FalkorDB graph database hostname
`FALKORDB_PORT`	Yes	`6379`	FalkorDB port
`FALKORDB_PASSWORD`	No	unset	FalkorDB/Redis password (set in production)
`FALKORDB_GRAPH`	No	`memories`	Graph name for Cypher queries
`GRAPH_NAME`	No	`memories`	Alias for `FALKORDB_GRAPH`
`PORT`	No	`8001`	Flask API server port

Vector Search (Qdrant)

AutoMem operates in graph-only mode if these variables are not set. Qdrant enables semantic vector search but is optional.

Variable	Required	Default	Description
`QDRANT_URL`	No	unset	Qdrant API endpoint (HTTP or HTTPS)
`QDRANT_API_KEY`	No	unset	Qdrant authentication key (required for Qdrant Cloud)
`QDRANT_COLLECTION`	No	`memories`	Collection name for memory vectors
`COLLECTION_NAME`	No	`memories`	Alias for `QDRANT_COLLECTION`
`VECTOR_SIZE`	No	`3072`	Embedding dimension — must match collection (768/1024/2048/3072)

Authentication

Variable	Required	Default	Description
`AUTOMEM_API_TOKEN`	Yes	unset	Token for all standard API operations
`ADMIN_API_TOKEN`	Yes	unset	Token for admin endpoints (`/admin/*`, `/enrichment/reprocess`)
`API_TOKEN`	No	unset	Backward-compatible alias for `AUTOMEM_API_TOKEN`
`ADMIN_TOKEN`	No	unset	Backward-compatible alias for `ADMIN_API_TOKEN`

See Authentication for token generation and usage details.

Embeddings

Variable	Required	Default	Description
`EMBEDDING_PROVIDER`	No	`auto`	Provider selection: `auto` / `voyage` / `openai` / `local` / `ollama` / `placeholder`
`EMBEDDING_MODEL`	No	`text-embedding-3-large`	OpenAI model name
`OPENAI_API_KEY`	No	unset	OpenAI (or compatible provider) API key
`OPENAI_BASE_URL`	No	unset	Custom base URL for OpenAI-compatible APIs (OpenRouter, LiteLLM, vLLM)
`VOYAGE_API_KEY`	No	unset	Voyage AI API key
`VOYAGE_MODEL`	No	`voyage-4`	Voyage model name (`voyage-4`, `voyage-4-large`)
`OLLAMA_BASE_URL`	No	`http://localhost:11434`	Ollama server endpoint
`OLLAMA_MODEL`	No	`nomic-embed-text`	Ollama embedding model
`EMBEDDING_BATCH_SIZE`	No	`20`	Items per batch API call (reduces costs 40–50%)
`EMBEDDING_BATCH_TIMEOUT_SECONDS`	No	`2.0`	Max wait before flushing a partial batch

Enrichment Pipeline

Background enrichment runs after each memory is stored — it generates similarity links, entity tags, and summaries.

Variable	Required	Default	Description
`ENRICHMENT_MAX_ATTEMPTS`	No	`3`	Retry limit before marking a job failed
`ENRICHMENT_SIMILARITY_LIMIT`	No	`5`	Number of semantic neighbors to link
`ENRICHMENT_SIMILARITY_THRESHOLD`	No	`0.8`	Min cosine similarity for `SIMILAR_TO` edge
`ENRICHMENT_IDLE_SLEEP_SECONDS`	No	`2`	Worker sleep duration when queue is empty
`ENRICHMENT_FAILURE_BACKOFF_SECONDS`	No	`5`	Delay between retry attempts
`ENRICHMENT_ENABLE_SUMMARIES`	No	`true`	Auto-generate memory summaries
`ENRICHMENT_SPACY_MODEL`	No	`en_core_web_sm`	spaCy model for NER (if installed)

Consolidation Engine

Background maintenance cycles that decay, cluster, and optionally forget low-value memories.

Variable	Required	Default	Description
`CONSOLIDATION_TICK_SECONDS`	No	`60`	Scheduler check interval
`CONSOLIDATION_DECAY_INTERVAL_SECONDS`	No	`86400`	Decay cycle frequency (1 day)
`CONSOLIDATION_DECAY_IMPORTANCE_THRESHOLD`	No	`0.3`	Min importance to process in decay
`CONSOLIDATION_CREATIVE_INTERVAL_SECONDS`	No	`604800`	Creative association cycle frequency (1 week)
`CONSOLIDATION_CLUSTER_INTERVAL_SECONDS`	No	`2592000`	Cluster pattern cycle frequency (1 month)
`CONSOLIDATION_FORGET_INTERVAL_SECONDS`	No	`0`	Forget cycle frequency (disabled by default)
`CONSOLIDATION_ARCHIVE_THRESHOLD`	No	`0.2`	Relevance threshold for archiving
`CONSOLIDATION_DELETE_THRESHOLD`	No	`0.05`	Relevance threshold for deletion
`CONSOLIDATION_GRACE_PERIOD_DAYS`	No	`30`	Min age in days before a memory can be forgotten
`CONSOLIDATION_IMPORTANCE_PROTECTION_THRESHOLD`	No	`0.7`	Memories above this importance are never forgotten
`CONSOLIDATION_PROTECTED_TYPES`	No	`Decision,Pattern`	Comma-separated types to never forget

Search and Recall

Variable	Required	Default	Description
`SEARCH_WEIGHT_VECTOR`	No	`0.25`	Vector similarity weight
`SEARCH_WEIGHT_KEYWORD`	No	`0.15`	Keyword/TF-IDF matching weight
`SEARCH_WEIGHT_TAG`	No	`0.10`	Tag overlap weight
`SEARCH_WEIGHT_IMPORTANCE`	No	`0.05`	User-assigned importance weight
`SEARCH_WEIGHT_RECENCY`	No	`0.10`	Freshness boost weight
`SEARCH_WEIGHT_CONFIDENCE`	No	`0.05`	Memory confidence weight
`SEARCH_WEIGHT_EXACT`	No	`0.25`	Content token overlap weight
`RECALL_MAX_LIMIT`	No	`100`	Maximum results returned by `/recall`
`RECALL_RELATION_LIMIT`	No	`5`	Max related memories per result
`RECALL_EXPANSION_LIMIT`	No	`20`	Max memories added via `expand_relations=true`

Sync Worker

Variable	Required	Default	Description
`SYNC_CHECK_INTERVAL_SECONDS`	No	`300`	Frequency of drift checks between FalkorDB and Qdrant (5 minutes)
`SYNC_AUTO_REPAIR`	No	`true`	Automatically queue missing embeddings when drift detected

Memory Types and Classification

Variable	Required	Default	Description
`MEMORY_TYPES`	No	`Decision,Pattern,Preference,Style,Habit,Insight,Context,Memory`	Comma-separated valid memory types
`RELATIONSHIP_TYPES`	No	`RELATES_TO,LEADS_TO,...`	Comma-separated valid relationship types
`ALLOWED_RELATIONS`	No	Same as `RELATIONSHIP_TYPES`	Alias for backward compatibility
`CLASSIFICATION_MODEL`	No	`gpt-4o-mini`	OpenAI model used for content classification fallback

API Server

Variable	Required	Default	Description
`LOG_LEVEL`	No	`INFO`	Python logging level (`DEBUG`, `INFO`, `WARNING`, `ERROR`)
`FLASK_ENV`	No	`production`	Flask environment mode

MCP Client Variables

These variables configure the mcp-automem client package, not the server.

Variable	Required	Default	Description
`AUTOMEM_ENDPOINT`	Yes	`http://127.0.0.1:8001`	HTTP URL of the AutoMem server
`AUTOMEM_API_KEY`	No	unset	API key for authenticated instances (preferred name)
`AUTOMEM_API_TOKEN`	No	unset	Alternative name for the API key
`AUTOMEM_LOG_LEVEL`	No	unset	Set to `debug` for verbose MCP client logging
`AUTOMEM_PROCESS_TAG`	No	unset	Process title tag for safe process management
`MCP_PROCESS_TAG`	No	unset	Alternative process tag variable

The client checks API key variables in this priority order: AUTOMEM_API_KEY → AUTOMEM_API_TOKEN → AUTOMEM_TOKEN → API_KEY.

Testing Variables

Used only for the AutoMem test suite — do not set in production.

Variable	Required	Default	Description
`AUTOMEM_RUN_INTEGRATION_TESTS`	No	`0`	Enable integration test suite
`AUTOMEM_START_DOCKER`	No	`0`	Auto-start Docker Compose before tests
`AUTOMEM_STOP_DOCKER`	No	`0`	Auto-stop Docker after tests
`AUTOMEM_TEST_BASE_URL`	No	`http://localhost:8001`	Test target URL
`AUTOMEM_ALLOW_LIVE`	No	`0`	Allow tests against non-localhost URLs
`AUTOMEM_TEST_API_TOKEN`	No	unset	API token for integration tests
`AUTOMEM_TEST_ADMIN_TOKEN`	No	unset	Admin token for integration tests

Example Configurations

Minimal local (graph-only, no vector search)

AUTOMEM_API_TOKEN=your-token-here
ADMIN_API_TOKEN=your-admin-token-here
FALKORDB_HOST=localhost
FALKORDB_PORT=6379
PORT=8001

Local with OpenAI embeddings and Qdrant

AUTOMEM_API_TOKEN=your-token-here
ADMIN_API_TOKEN=your-admin-token-here
FALKORDB_HOST=localhost
PORT=8001
OPENAI_API_KEY=sk-...
QDRANT_URL=http://localhost:6333
VECTOR_SIZE=3072

Local with Ollama (fully offline)

AUTOMEM_API_TOKEN=your-token-here
ADMIN_API_TOKEN=your-admin-token-here
FALKORDB_HOST=localhost
PORT=8001
EMBEDDING_PROVIDER=ollama
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=nomic-embed-text
QDRANT_URL=http://localhost:6333
VECTOR_SIZE=768

Railway production

PORT=8001
FALKORDB_HOST=falkordb.railway.internal
FALKORDB_PORT=6379
FALKORDB_PASSWORD=<generated>
AUTOMEM_API_TOKEN=<generated>
ADMIN_API_TOKEN=<generated>
OPENAI_API_KEY=sk-...
QDRANT_URL=https://your-cluster.cloud.qdrant.io
QDRANT_API_KEY=your-qdrant-key
VECTOR_SIZE=3072

MCP client (in platform config file)

{
  "env": {
    "AUTOMEM_ENDPOINT": "https://your-service.railway.app",
    "AUTOMEM_API_KEY": "your-api-token"
  }
}

Security Checklist

Never commit .env files — add to .gitignore
Use strong tokens — minimum 32 bytes: openssl rand -hex 32
Use separate tokens for AUTOMEM_API_TOKEN and ADMIN_API_TOKEN
Set FALKORDB_PASSWORD in all production environments
Use HTTPS for all external service endpoints (Qdrant Cloud, OpenAI, Voyage)
Use Railway private networking — never expose FalkorDB publicly
Avoid query parameter auth (?api_key=) in production — tokens appear in server logs