Filters

Praxis AI registers AI-specific filters into the Praxis filter pipeline. The generated filter reference is the authoritative inventory of filter names, descriptions, and configuration documentation.

For pipeline execution, filter traits, body access, conditional execution, filter chains, and core filters, see the Praxis core filter documentation.

Organization

AI filters are organized across two crates:

apis/src/                 Provider API integrations
  anthropic/              Anthropic Messages API
  openai/                 OpenAI Responses and Conversations APIs
  classifier/             Shared request classification
  store/                  Response and conversation persistence

filters/src/              Cross-provider behavior
  agentic/                MCP and A2A
  guardrails/             AI content guardrails
  inference/              Model routing
  prompt_enrich/          Prompt injection
  token_usage/            Token counting and headers

Praxis AI also inherits all base proxy filters from Praxis core through FilterRegistry::with_builtins(). This includes routing, load balancing, headers, credential injection, JSON-RPC parsing, CORS, compression, IP ACLs, and other general proxy behavior. Those filters are intentionally documented in the Praxis core filter reference, not duplicated here.

Registration

In-tree AI filters are registered by praxis_ai_filters::register_ai_filters. Downstream consumers that only need AI filters (for example an Envoy ExtProc) can depend on praxis-ai-filters without the proxy crate:

use praxis_filter::FilterRegistry;

let mut registry = FilterRegistry::with_builtins();
praxis_ai_filters::register_ai_filters(&mut registry);
// Or: let registry = praxis_ai_filters::build_ai_registry();

praxis-ai-proxy builds the full registry in three ownership layers:

let mut registry = FilterRegistry::with_builtins();
praxis_ai_filters::register_ai_filters(&mut registry);
register_external_filters(&mut registry); // proxy-only

This keeps core filters, in-tree AI filters, and auto-discovered extensions at clear ownership boundaries. External filter auto-discovery stays proxy-only.

Pipelines that use OpenAI store or rehydrate filters must also install the response-store extension:

pipeline.add_pipeline_extension(
    Box::new(praxis_ai_apis::store::ResponseStoreRegistry::new()),
);

The AI proxy does this in server/src/pipelines.rs. Other hosts (such as ExtProc) must do the same.

Related documentation

• Generated filter reference
• Feature overview
• Example configurations
• Writing extensions