Skip to content
For Platform Engineers

Production primitives, not a black box.

Cordum gives you NATS, Redis, gRPC, and a typed protocol—so you run agents on infrastructure you already know how to operate.

The Problem

Agent infra shouldn’t be a science project.

You need predictable dispatch, observable state, and controls that work under load.

Agent infra is a patchwork of queues, cron jobs, and hope
No standard wire format—every team invents its own job protocol
Retries, timeouts, and dead-letter handling are afterthoughts
Scaling agents means scaling the problem of uncontrolled execution
Architecture

Built on infra you already trust.

CAP v2 wire protocol

Protobuf-defined types (BusPacket, JobRequest, JobResult) give every service a shared contract. Go, Python, Node, and C++ SDKs included.

NATS bus + Redis state

NATS handles pub/sub and optional JetStream durability. Redis stores job state, context windows, config, and workflow runs. Battle-tested primitives, not a custom database.

gRPC Safety Kernel

Check, Evaluate, Explain, and Simulate over gRPC. Policy decisions are cached and reloadable from versioned bundles. Runs at :50051, horizontally scalable.

Operational controls

Routing rules, configurable retries, dead-letter queues, stale-job reconciliation, and pending replay. The Scheduler owns the full dispatch lifecycle.

Deploy agents the way you deploy services.

Cordum runs as a set of Go binaries communicating over NATS. No proprietary runtime, no vendor SDK lock-in.