Environment Variables¶
Neuralgentics uses a multi-layered environment configuration to allow seamless transition from local development to production clouds.
⛓️ The Inheritance Chain¶
Configuration is not read from a single place. It follows a strict precedence order where the right-most source always wins.
lowest trust ◄───────────────────────────────────────────────── highest trust
┌────────────┐ ┌────────────┐ ┌───────────┐ ┌───────────┐ ┌─────────┐
│ SYSTEM ENV │ ──► │ SHELL RC │ ──► │ .env │ ──► │ DIRENV │ ──► │ CLI │
└────────────┘ └────────────┘ └───────────┘ └───────────┘ └─────────┘
(Global) (/etc/bashrc) (Project) (Local) (Runtime)
Diagram 5 — Env Var Inheritance Chain. This sequence defines how Neuralgentics resolves configuration. If
OLLAMA_API_KEYis defined in both the.envfile and as a CLI flag, the CLI flag takes precedence.
📋 Variable Reference¶
Core Runtime (NEURALGENTICS_*)¶
| Variable | Default | Description |
|---|---|---|
NEURALGENTICS_PREFIX | ~/.neuralgentics | Installation root for binaries. |
NEURALGENTICS_API_KEY | None | Authentication key for the Go backend. |
NEURALGENTICS_LOG_LEVEL | info | debug, info, warn, error. |
Memory Engine (MEMINI_*)¶
| Variable | Default | Description |
|---|---|---|
MEMINI_DB_URL | postgres://... | Connection string for PostgreSQL + pgvector. |
MEMINI_EMBEDDING_MODE | auto | cpu (384), gpu (1024), or auto (dual). |
THOUGHT_CHAINS | true | Enable/disable sequential reasoning logs. |
MEMINI_SVR_PORT | 8000 | Port for the Python memory server. |
LLM Infrastructure (OLLAMA_*)¶
| Variable | Default | Description |
|---|---|---|
OLLAMA_API_KEY | None | Key for Ollama Cloud. |
OLLAMA_BASE_URL | https://ollama.com/v1 | API endpoint for model requests. |
OLLAMA_DEFAULT_MODEL | gemma4:31b | Fallback model if no specialist is defined. |
🛠️ Practical Usage¶
Setting a Local Override¶
To use a custom prefix without modifying your shell:
Managing Multiple Profiles¶
We recommend using direnv for project-specific memory servers: