Expand description
OpenAI-compatible AI privacy gateway that embeds redact-core in-process.
The gateway sits between an application and a model provider. On the way out it detects sensitive values and applies the action a policy profile specifies for each entity type: allow, block, mask, replace, hash, or tokenize. On the way back it can restore tokenized values so the caller sees a complete answer while the provider only ever saw placeholders.
use redact_gateway::{config::ResolvedConfig, GatewayServer};
let config = ResolvedConfig::load()?;
GatewayServer::new(config).await?.run().await§Subsystems
| Module | Responsibility |
|---|---|
config | Resolved configuration from files, environment and flags |
policy | Per-entity actions and profiles |
packs | Pattern packs loaded from YAML at runtime |
redact | Detection, action application, tokens and payload traversal |
vault | Pluggable token map backends |
auth | Inbound API key and OIDC authentication |
telemetry | OpenTelemetry traces, metrics and logs |
audit | Audit records emitted as OpenTelemetry log records |
Re-exports§
pub use config::ConfigError;pub use config::ConfigHandle;pub use config::ResolvedConfig;pub use error::GatewayError;pub use policy::EntityAction;pub use policy::PolicySet;pub use policy::Profile;pub use server::GatewayServer;
Modules§
- audit
- Audit emission as an OpenTelemetry Logs concern.
- auth
- Inbound authentication for the gateway.
- config
- Resolved gateway configuration and the sources that produce it.
- error
- Gateway error type rendered as OpenAI-compatible error payloads.
- openai
- packs
- Runtime loading of YAML pattern packs into a
redact_core::Recognizer. - policy
- Redaction policy: per-entity actions evaluated on the request hot path.
- proxy
- HTTP client for the OpenAI-compatible inference provider.
- redact
- Policy-driven redaction of text and OpenAI payloads.
- routes
- HTTP surface: OpenAI-compatible endpoints plus gateway utilities.
- server
- Server assembly: engine, pattern packs, token map, auth, audit and router.
- stream
- Server-sent event handling for streaming chat completions.
- telemetry
- OpenTelemetry SDK bootstrap for traces, metrics and logs.
- vault
- Pluggable token-map backends.