Essential settings
Set server auth, embedding, View LLM, Redis cache, and backup defaults used by daily operation.
Use this page as the checklist for the settings that make Lore usable after installation.
There are two kinds of settings:
| Kind | Where to set it | Examples |
|---|---|---|
| Deployment settings | Server environment or Docker Compose .env | database URL, API token, Redis, storage paths |
| Runtime settings | /setup, /settings, or /api/settings | embedding model, View LLM, recall, Dream, backup policy |
A good rule: set infrastructure values before the server starts, then use the Web Console for model and memory behavior.
Required server setting
| Setting | Set it in | What it does |
|---|---|---|
DATABASE_URL | Web server environment | Connects Lore Web to PostgreSQL. The server cannot start correctly without it. |
For Docker Compose, DATABASE_URL usually points at the bundled postgres service. For managed PostgreSQL, point it at your own host.
Recommended server settings
| Setting | Set it when | Notes |
|---|---|---|
API_TOKEN | The server is reachable by anything you do not fully trust | Use the same token in the Web UI, MCP clients, and hooks. |
REDIS_URL | You run more than one server process or want shared cache | Lore can run without Redis, but Redis is better for shared deployments. |
CACHE_KEY_PREFIX | Redis is shared with other apps | Keeps Lore keys separate. |
CACHE_LOCAL_MAX_ITEMS | You are not using Redis | Limits the in-process fallback cache. |
LORE_DB_POOL_SIZE | You need to tune database connections | Default is 10; keep it small unless you know you need more. |
Docker Compose deployments also use host-side paths such as POSTGRES_DATA_DIR, REDIS_DATA_DIR, and SNAPSHOT_DATA_DIR so data survives container restarts.
Runtime model settings
Configure these in /setup first, then adjust them later in /settings:
| Setting group | Required for | Recommendation |
|---|---|---|
| Embedding provider, base URL, API key, model | Semantic recall and index rebuilds | Use an OpenAI-compatible embedding endpoint. |
| View LLM provider, base URL, API key, model | Generated views and Dream | Use a model that follows structured instructions reliably. |
| Recall settings | What agents see in recall blocks | Start with defaults; tune after you have real memory content. |
| Dream settings | Memory maintenance suggestions | Keep manual review enabled until you trust the workflow. |
| Backup settings | Safe cleanup and migration | Configure before large edits or Dream runs. |
Quick diagnosis
| Symptom | Check first |
|---|---|
| Server does not start | docker compose ps and docker compose logs web from the directory with docker-compose.yml |
| Web Console works but MCP fails | API_TOKEN, MCP URL, client_type, agent environment |
| Recall returns weak results | memory content, disclosure, glossary, then recall thresholds |
| Dream cannot generate useful suggestions | View LLM settings and memory quality |