Skip to main content

AI Cost Firewall v0.4.1

AI Cost Firewall v0.4.1 introduced the first production wiring for asynchronous structured-evidence delivery to VCAL Audit.

Highlights​

  • initial buffered HTTP evidence sink
  • configurable Audit URL, API key, and producer instance ID
  • bounded in-memory queue and batch delivery
  • flush interval, request timeout, retry, and backoff controls
  • non-blocking evidence delivery on the normal LLM request path
  • continued VCAL Privacy Guard and VCAL Security Guard orchestration

VCAL Audit delivery​

Example configuration for the initial integration:

audit_enabled true;
audit_url http://vcal-audit:8092;
audit_api_key replace-with-shared-audit-token;
audit_producer_instance_id ai-firewall-01;
audit_queue_capacity 10000;
audit_batch_size 100;
audit_flush_interval_ms 1000;
audit_request_timeout_seconds 5;
audit_retry_max_attempts 5;
audit_retry_initial_backoff_ms 250;

The sender is memory-backed. Temporary failures are retried, but a batch that remains undeliverable after retry exhaustion is dropped and logged rather than replayed from durable storage.

Next release​

Version 0.4.2 hardens the evidence lifecycle contract, finalizes schema version 1.1, adds explicit guard event names, preserves Security Guard rule identifiers, and enforces exactly one terminal event per accepted trace.