Configuration Reference

AllCodex services use environment variables for configuration. The following tables list the available variables.

AllCodex Core (database)

These variables configure database paths, ports, and startup credentials.

Variable	Default	Description
`PORT`	`8080`	The port the Express HTTP server binds to on the host.
`TRILIUM_DATA_DIR`	`/srv/allcodex/core-data`	Directory where SQLite files, backups, and configurations are stored.
`ALLCODEX_PASSWORD`	(none)	Startup password for the database. Required for zero-login auto-provisioning.
`USER_UID`	`999`	UID to run the server process as inside the container.
`USER_GID`	`988`	GID to run the server process as inside the container.

AllKnower (parsing brain)

These variables configure database models, connections, OpenRouter keys, and model selections.

System settings

Variable	Default	Description
`PORT`	`3001`	The port the Elysia server binds to.
`DATABASE_URL`	(none)	PostgreSQL connection URI for Prisma (accounts, history, and budgets).
`LANCE_DB_PATH`	`/srv/allcodex/knower-data/lancedb`	Directory where LanceDB vector and FTS tables are persisted.
`OPENROUTER_API_KEY`	(none)	Your OpenRouter API Key.
`OPENROUTER_BASE_URL`	`https://openrouter.ai/api/v1`	Custom OpenRouter API base endpoint.
`LOCAL_PROVIDER_BASE_URL`	`http://localhost:11434/v1`	OpenAI-compatible base URL for a local model server (Ollama, LM Studio, etc.).
`LOCAL_PROVIDER_API_KEY`	`ollama`	API key for the local provider. Most local servers accept any non-empty value.
`PORTAL_INTERNAL_SECRET`	(none)	A secure, random token shared with the Portal to verify auto-provisioning.
`BETTER_AUTH_SECRET`	(none)	Secret used by better-auth to sign cookies and tokens. Must be $\ge 16$ characters in production.
`BETTER_AUTH_URL`	`http://localhost:3001`	The canonical URL of the AllKnower auth server.
`INTEGRATION_CREDENTIALS_KEY`	(none)	Critical in production. A 32-byte key (represented as a 64-character hex string) that encrypts database credentials at rest.

Web push and email notifications

Variable	Default	Description
`VAPID_PUBLIC_KEY`	(none)	VAPID public key for Web Push notifications.
`VAPID_PRIVATE_KEY`	(none)	VAPID private key for Web Push notifications.
`VAPID_SUBJECT`	`mailto:admin@example.com`	Notification sender contact email address.
`RESEND_API_KEY`	(none)	Resend email delivery API key.
`RESEND_FROM_EMAIL`	`noreply@allcodex.local`	The email address from which budget alerts are sent.

Override the default models for specific tasks. Each task supports a primary model and up to three fallback models.

Task	Variable	Default Model
Brain Dump	`BRAIN_DUMP_MODEL`	`deepseek/deepseek-v4-pro`
Consistency Checker	`CONSISTENCY_MODEL`	`deepseek/deepseek-v4-pro`
Relationship Suggester	`SUGGEST_MODEL`	`google/gemini-3.1-flash-lite`
Gap Detector	`GAP_DETECT_MODEL`	`deepseek/deepseek-v4-pro`
Autocomplete	`AUTOCOMPLETE_MODEL`	`openai/gpt-5-nano`
Embeddings	`EMBEDDING_CLOUD`	`qwen/qwen3-embedding-8b`
Rerank	`RERANK_MODEL`	`cohere/rerank-4-pro`

Fallback syntax: Append _FALLBACK_1, _FALLBACK_2, or _FALLBACK_3 to any model key (e.g. BRAIN_DUMP_MODEL_FALLBACK_1=google/gemini-2.5-pro).
Auto route: Set USE_OPENROUTER_AUTO=true to let OpenRouter automatically select the best model for all tasks.

Local model providers

AllKnower can route any parser or embedding task to a local OpenAI-compatible server such as Ollama, LM Studio, or vLLM. This lets you run AllCodex fully offline or mix local and cloud models in the same fallback chain. To send a task to your local provider, prefix the model name with ollama/ or local/. AllKnower strips the prefix before calling the local server and automatically removes OpenRouter-only fields (models, plugins, provider) so requests stay compatible. Local models participate in the same fallback chain as cloud models. You can lead with a local model and fall back to a cloud model when the local server is offline, or the reverse.

# Run brain dump locally, fall back to OpenRouter if Ollama is down
BRAIN_DUMP_MODEL=ollama/llama3
BRAIN_DUMP_FALLBACK_1=deepseek/deepseek-v4-pro

# Embed locally with nomic-embed-text (768 dimensions)
EMBEDDING_CLOUD=ollama/nomic-embed-text
EMBEDDING_DIMENSIONS=768

The LanceDB vector table is created with a fixed dimension. If you change EMBEDDING_CLOUD to a model with a different output size (for example switching from a 1024-dim cloud model to a 768-dim local model), set EMBEDDING_DIMENSIONS to match, delete the existing LanceDB table, and run POST /rag/reindex to rebuild it.

AllCodex Portal (frontend)

These variables configure connections to backend services and secret handshakes.

Variable	Default	Description
`PORT`	`3000`	Port the Next.js production server runs on.
`ALLCODEX_URL`	`http://localhost:8080`	The private network URL of the running AllCodex Core container.
`ALLKNOWER_URL`	`http://localhost:3001`	The private network URL of the running AllKnower container.
`PORTAL_INTERNAL_SECRET`	(none)	Critical. Must match `PORTAL_INTERNAL_SECRET` in AllKnower to authorize session cookies.
`NEXT_PUBLIC_PORTAL_URL`	`http://localhost:3000`	The public URL of the portal, used to generate player-facing share links.
`LORE_ROOT_NOTE_ID`	(none)	System fallback default root lore note ID.
`ALLCODEX_ETAPI_TOKEN`	(none)	System fallback token for Core database connections.
`ALLKNOWER_BEARER_TOKEN`	(none)	System fallback token for AllKnower communication.
`VAPID_PUBLIC_KEY`	(none)	Public key for push notification client subscriptions.

Getting Started

Concepts

Guides

Reference

Self-Hosting

Contributing

Changelog

Configuration Reference

AllCodex Core (database)

AllKnower (parsing brain)

System settings

Web push and email notifications

Local model providers

AllCodex Portal (frontend)

​AllCodex Core (database)

​AllKnower (parsing brain)

​System settings

​Web push and email notifications

​Local model providers

​AllCodex Portal (frontend)

AllCodex Core (database)

AllKnower (parsing brain)

System settings

Web push and email notifications

Local model providers

AllCodex Portal (frontend)