Configuration
All configuration is via environment variables. Set them in ~/.abmind/config/.env.memory or export them before running abmind commands.
Paths
| Variable | Default | Description |
|---|---|---|
ABMIND_HOME | ~/.abmind | Root directory for all abmind data |
MEMORY_DIR | $ABMIND_HOME/memory | Directory containing memory.db and consolidation files |
ABMIND_KEY_FILE | $ABMIND_HOME/secret/abmind.key | Encryption key for class-3 secrets |
Memory
| Variable | Default | Description |
|---|---|---|
MEMORY_SEARCH_MODE | hybrid | Search mode: hybrid, fts, signature |
MEMORY_MAX_DB_SIZE_MB | 4096 | DB size threshold — triggers aging acceleration warning |
MEMORY_ORIGINAL_TTL_DAYS | 90 | Days before original-language content expires |
MEMORY_ENGLISH_TTL_DAYS | 14 | Days before English translation expires (if original preserved) |
MEMORY_AGING_ENABLED | true | Enable/disable automatic memory aging |
MEMORY_IPC | 1 | Enable IPC notifications between processes (set 0 to disable) |
MEMORY_BACKEND | sqlite | Storage backend (only sqlite currently supported) |
Embeddings
| Variable | Default | Description |
|---|---|---|
EMBEDDING_ENABLED | false | Enable vector embeddings (requires Ollama or OpenAI) |
EMBEDDING_PROVIDER | ollama | Provider: ollama or openai |
EMBEDDING_MODEL | nomic-embed-text | Model name for embedding generation |
EMBEDDING_URL | http://localhost:11434 | Ollama API endpoint |
EMBEDDING_API_KEY | (empty) | API key (required for openai provider) |
EMBEDDING_DIMENSIONS | 768 | Vector dimensions (must match model output) |
EMBEDDING_SIMILARITY_THRESHOLD | 0.5 | Minimum cosine similarity for Se stage results (0.0–1.0) |
System One judgments (#1812)
| Variable | Default | Description |
|---|---|---|
SYSTEM1 | laya | Judgment backend: laya (local sidecar, silent fallback), jev (SaaS, explicit egress opt-in), or off |
SYSTEM1_RECALL | on | off disables the recall rerank; default on since #1813, explicit off stays authoritative |
SYSTEM1_FASTPATH | off | on enables fast-path decisions with passing profiles only (repeat/attribution; lookup bypass has none) — never SaaS permission |
SYSTEM1_JEV_EGRESS | `` | Comma-separated per-operation Jev egress grants (rerank,lookup,repeat,attribution); empty means no SaaS payloads |
SYSTEM1_TIMEOUT_MS | 1500 | Per-request timeout in ms (100–10000) |
SYSTEM1_MAX_CANDIDATES | 20 | Max recall candidates judged per batch (2–20) |
JEV_URL | https://api.typesafe.ai | Jev base URL (must stay bare https) |
JEV_API_KEY | (empty) | Jev API key — required for SYSTEM1=jev, never logged |
JEV_MODEL | jev-1.13.0 | Pinned Jev model (never a latest alias) |
LAYA_URL | http://127.0.0.1:8765 | Local Laya sidecar endpoint (must stay bare loopback) |
See System One judgments for setup, the sidecar, and privacy semantics.
Recall tuning
| Variable | Default | Description |
|---|---|---|
RECALL_DECAY_DAYS | 365 | Days until recency factor reaches floor |
RECALL_DECAY_FLOOR | 0.3 | Minimum recency multiplier (0.0–1.0) |
RECALL_EMOTION_BOOST | 0.1 | Emotion score multiplier for recency boost |
SIGNATURE_BITS | 256 | Bit width for locality-sensitive hash signatures |
RECALL_CONTEXT_HOUR_BOOST | 0.02 | Score boost for same time-of-day bucket |
RECALL_CONTEXT_DOW_BOOST | 0.015 | Score boost for same day-of-week |
RECALL_CONTEXT_TOPIC_BOOST | 0.03 | Score boost for matching conversation topic |
ABM-L compression
| Variable | Default | Description |
|---|---|---|
ABML_VERSION | plain | Compression version: plain, v0, v1, v2 |
ABML_MIN_CHARS | 100 | Minimum content length to apply compression |
Context tiers (#348)
| Variable | Default | Description |
|---|---|---|
CONTEXT_TIER_ENABLED | true | Enable three-tier context assembly |
CONTEXT_TIER_TAIL | 20 | Number of recent messages in tail tier |
CONTEXT_TIER_MIDDLE | 50 | Number of messages in middle tier |
COMPACTION_LLM_ENABLED | false | Use LLM for context compaction (expensive) |
Sleep cycle
| Variable | Default | Description |
|---|---|---|
SLEEP_QUALITY | (unset) | Sleep depth: basic, budget, normal, ultimate |
SLEEP_TIMEOUT_MIN | 55 | Wall-clock timeout in minutes |
SLEEP_MAX_LLM_CALLS | 18 | Maximum LLM calls per sleep cycle (hard cap: 50) |
SLEEP_CURATION_DAY | sunday | Day of week for weekly curation tasks |
SLEEP_MODEL_NAME | unknown | Model identifier for audit logging |
AGENT_SLEEP_CTX_WINDOW | 128000 | Context window size for sleep LLM calls |
Security
| Variable | Default | Description |
|---|---|---|
ABTARS_PASS | (unset) | Passphrase for secret encryption/decryption |
ABMIND_USER | (unset) | Current user identity for multi-user isolation |
File locations
~/.abmind/
├── config/
│ ├── .env.memory ← env vars go here
│ └── users.json ← multi-user config
├── memory/
│ ├── memory.db ← SQLite database
│ ├── consolidation/
│ │ ├── daily/
│ │ ├── weekly/
│ │ └── quarterly/
│ ├── working/ ← daily transcript scratch
│ ├── sleep/ ← sleep audit logs
│ ├── core/ ← core identity templates
│ └── todo.md ← persistent todo list
├── secret/
│ └── abmind.key ← encryption key (chmod 600)
├── backups/ ← encrypted backups
└── topics/ ← topic knowledge filesOverriding for development
For testing or development, override paths:
bash
export ABMIND_HOME=/tmp/abmind-test
export MEMORY_DIR=/tmp/abmind-test/memory
abmind install # creates structure in /tmpAll env vars are parsed once at startup via getAbmindEnv(). Changes require process restart.