Environment variables configure both the InferenceWall server and SDK behavior. Set them in your shell, pass them to Docker with -e, or load them from .env.local via source .env.local before running inferwall serve.
Variable Reference
| Variable | Description | Default |
|---|
IW_API_KEY | Scan API key. Required for authenticated scan requests. | None (dev mode — auth disabled) |
IW_ADMIN_KEY | Admin API key. Required for admin endpoints. | None (dev mode — auth disabled) |
IW_HOST | Server bind address. | 0.0.0.0 |
IW_PORT | Server port. | 8000 |
IW_TLS | TLS mode: auto, off, or acme. | off |
IW_PROFILE | Detection profile: lite, standard, or full. | lite |
IW_LOG_LEVEL | Log level: debug, info, warning, or error. | info |
IW_REDIS_URL | Redis connection URL for distributed session storage. | None (in-memory) |
IW_SIGNATURES_DIR | Path to a directory of custom signature YAML files, merged with the built-in catalog at startup. | ~/.inferwall/signatures |
IW_POLICY_PATH | Path to a custom policy YAML file. | Auto-discovery from ~/.inferwall/policies/ |
IW_ELK_URL | Logstash HTTP endpoint for shipping logs to an ELK stack. Requires pip install inferwall[observability]. | None (disabled) |
Details
Authentication keys (IW_API_KEY, IW_ADMIN_KEY)
InferenceWall uses two separate keys with distinct roles:
IW_API_KEY — authorizes scan requests (/v1/scan/input, /v1/scan/output). Include it as Authorization: Bearer <key> in HTTP requests.IW_ADMIN_KEY — authorizes admin endpoints. Keep this key restricted to operators.
When neither key is set, authentication is completely disabled. This is intended for local development only — never run without keys in production.
Detection profile (IW_PROFILE)
The profile controls which detection engines are active and which signatures are evaluated:
| Value | Engines | Signatures used | Latency |
|---|
lite | Heuristic (Rust) | 75 heuristic | <0.3 ms p99 |
standard | + Classifier (ONNX) + Semantic (FAISS) | + 11 classifier + 10 semantic | <80 ms p99 |
full | + LLM-Judge | + composite (ambiguous inputs only) | <2 s p99 |
TLS mode (IW_TLS)
| Value | Behavior |
|---|
off | Plain HTTP (default) |
auto | TLS with automatic certificate management |
acme | TLS via ACME/Let’s Encrypt |
Redis (IW_REDIS_URL)
Set IW_REDIS_URL to a Redis connection string (e.g., redis://localhost:6379) when running multiple InferenceWall instances that need to share session state. Without it, sessions are stored in process memory.
Custom signatures (IW_SIGNATURES_DIR)
By default, InferenceWall loads custom signatures from ~/.inferwall/signatures/. All .yaml files in that directory are merged with the built-in catalog at startup. A custom signature with the same id as a built-in signature replaces it entirely.
Set IW_SIGNATURES_DIR to point at a different directory:
export IW_SIGNATURES_DIR=/opt/inferwall/team-signatures
Custom policy (IW_POLICY_PATH)
By default, InferenceWall auto-discovers policy files from ~/.inferwall/policies/. Set IW_POLICY_PATH to load a specific policy file instead:
export IW_POLICY_PATH=~/.inferwall/policies/my-policy.yaml
ELK observability (IW_ELK_URL)
Set this to your Logstash HTTP input endpoint to ship scan events and server logs to an ELK stack. Requires the observability extra:
pip install inferwall[observability]
export IW_ELK_URL=http://logstash.internal:5044
Generating keys
Use inferwall admin setup to generate both keys and write them to .env.local:
inferwall admin setup
source .env.local && inferwall serve
Add .env.local to your .gitignore immediately. It contains secrets and must never be committed to source control.
Docker example
docker run -p 8000:8000 \
-e IW_API_KEY=iwk_scan_yourkey \
-e IW_ADMIN_KEY=iwk_admin_yourkey \
-e IW_PROFILE=standard \
inferwall
