The compounding loop, the agents, and the trust contract.
Most reliability tools are a feature. Causum is a loop. Domain → Signal → RCA → Remediation → feedback to Domain. Below: how each layer talks to the next, what runs in the hot path vs the judgment path, and how the trust contract is enforced.
Domain → Signal → RCA → Remediation → feedback to Domain. The loop compounds with every incident.
Two pipelines. Different latency budgets. Same guardrails.
🔥 Hot-path pipeline
Sensors, alerts, and deterministic guardrails. Operates at the speed of your incident. No LLMs in this path — predictable, fast, auditable.
Latency budget: milliseconds to seconds.
🤖 Agent runtime
The Codebase Explorer, Diagnosis, Remediation, and Briefing agents. Each has bounded scope, named purpose, same guardrails. Their judgment feeds the hot path; their reasoning is always evidence-backed.
Latency budget: seconds to minutes.
Three modes. Same product. Different isolation.
- ☁️
Causum Cloud (multi-tenant)
Causum-managed, multi-tenant. Lowest operational burden; non-regulated workloads.
- 🏢
Single-tenant Cloud
Dedicated Causum-managed environment per customer. Customer isolation requirements.
- 🔒
VPC-bound (Customer Cloud)
Inside your AWS / Azure / GCP VPC. Data residency, air-gapped, regulated workloads.
Causum is engineered for the assumption that it might be wrong.
The autonomy ladder — Observe → Assist → Supervised → Autopilot — applies per service. Promotion is explicit. Auto-demotion on failure. Six guardrails on every action. Kill switch always visible.
Every action is signed and chained into a cryptographic audit log. SIEM-streamable. Forever-queryable.
Want the long-form architecture?
The architecture overview PDF (24 pages) covers every component, data flow, and trust pattern in detail. We share it with prospective customers under NDA.