Runtime Model

BlazeRules has one rule engine and several ways to feed it.

Runtime	What it is	Use it when
`blazerules` Python module	Direct Python binding to the engine.	Applications that already control batching or use PyArrow/JSON.
C++ library	Native API for embedding in a service.	In-process integration with a C++ service.
`blazerules` CLI	Native batch/stream command using the core and IO stack.	Validation, evaluation, backtesting, and streaming without Python code.
`blazerules_io`	Kafka, binary decoders, file readers, CDC helpers.	Broker, file, and format adapters with application-owned control flow.
`blazerules_agent`	Local ingest binary for HTTP, stdin, and file tails.	A small local collector without application code.
`blazerules_dashboard`	Read-only UI over decision/DLQ/metrics/result files.	Local inspection of decisions, models, rules, and failures.
Helm chart	Agent DaemonSet plus dashboard Deployment.	Node-local Kubernetes log collection.

Embedded Python

import blazerules

engine = blazerules.RuleEngine()
engine.load_rules("rules.yaml")
result = engine.evaluate_ndjson(payload)

This mode fits applications that already hold events in memory and can call the engine once per batch.

IO Module

import blazerules_io

payload = blazerules_io.read_ndjson_bytes("s3://bucket/events/day.ndjson")

This mode provides file, Kafka, CDC, and binary decoding helpers while leaving control flow with the application.

Native CLI

blazerules eval --rules rules.yaml --input arrow-ipc --path events.arrow --output grouped-decisions
blazerules stream kafka --rules rules.yaml --brokers localhost:9092 --input-topic transactions --format protobuf --descriptor schema.desc --message payments.Transaction

This mode exposes the same data-plane capability as the Python package through files, stdin, S3 objects, or Kafka topics. Python-only in-memory objects such as pyarrow.RecordBatch map to CLI transports such as Arrow IPC, Parquet, or CSV.

Agent

blazerules_agent --rules rules.yaml --input http --port 9480 --output ndjson --output-path decisions.ndjson

Use the agent for local log-style ingestion: HTTP /v1/logs, stdin, or file tails. The agent still evaluates in batches; it is not a per-record rule server.

Dashboard

blazerules_dashboard \
  --decision-log decisions.ndjson \
  --dead-letter-log dead_letters.ndjson \
  --rules rules.yaml

The dashboard is read-only. It does not mutate rules or engine state.

Kubernetes

The Helm chart runs the agent as a DaemonSet and the dashboard as a Deployment. The default mode tails container logs from each node and writes decision/DLQ files under /var/log/blazerules.

See Kubernetes Log Ingestion and Deployment.