Seven tools. One layer.

Rauha is a runtime isolation layer for AI agents. On Linux it enforces policies at the kernel level via eBPF LSM — syscall filtering, filesystem boundaries, network namespacing — with no hypervisor required. On macOS it uses Virtualization.framework for equivalent containment. Agents operate freely within their defined scope and cannot touch anything outside it.

Ahti is Git for infrastructure. Every resource — pod, node, deployment, service — gets a version history where each version points to the event that caused it. Causality isn't computed after the fact; it's structural. By month six, Ahti knows that the Tuesday 3am batch job causes memory pressure on node-7, which cascades to three pod restarts. No rules configured. It learned. Query via ahti blame, ahti diff, ahti bisect, or ahti root-cause — the same interface for humans and agents alike.

Sykli is a CI orchestration system built for AI-native workflows. Pipelines are code — not YAML — so they can branch, reason, and adapt based on what the agent sees. Built on the BEAM for native fault-tolerance and concurrency. When a step fails, Sykli understands the failure in context rather than stopping blindly.

Kerto is a local context keeper for AI agents. It maintains a knowledge graph of the agent's operational environment — services, dependencies, recent actions, known states — and applies decay so stale information fades rather than persists. Agents that query Kerto get a picture of the world that reflects now, not three deployments ago.

Nopea is a deployment tool with memory. It builds a knowledge graph across every deploy so each one makes the next smarter. Strategy is selected automatically based on history — a service that has failed under concurrent deploys gets a canary rollout, without anyone writing a rule. Integrates with Kerto, emits False Protocol events, and exposes a CLI, HTTP API, and MCP server so humans and agents use the same interface.

Polku is a high-performance event gateway written in Rust. It ingests agent event streams over gRPC and routes them through a composable plugin architecture — filter, enrich, transform, forward. All False Protocol events enter the stack through Polku. Plugins load at runtime and compose arbitrarily.

False Protocol is an open event standard for AI-native infrastructure. Conventional structured logging defines fields like level and message. False Protocol defines fields that carry meaning for automated reasoning: what_failed, why_it_matters, possible_causes, affected_scope. It's not tied to any False Systems tool — any agent, runtime, or observability system can implement it.