Fathom
v0.4 · OPEN SOURCEA CLIPS-based expert system for AI agent governance. Define rules in natural patterns, enforce them at runtime.
rules: [{ when: "agent.clearance < data.classification", then: "deny" }] Deep dive →
Kraken Networks builds AI governance, service mesh, and orchestration tools.
Open source, built for depth.
PRODUCTS
A CLIPS-based expert system for AI agent governance. Define rules in natural patterns, enforce them at runtime.
rules: [{ when: "agent.clearance < data.classification", then: "deny" }] Deep dive → Intelligent service mesh for AI agent networks. Route, observe, and control inter-agent communication.
mesh.route({ from: "analyst", to: "classifier", policy: "least-privilege" }) Deep dive → Workflow orchestration for multi-agent systems. Define pipelines, manage state, handle failures gracefully.
pipeline.stage("review", { agents: 3, consensus: "majority" }) Deep dive → 20+ tools built / Herndon, VA
ARCHITECTURE
Kraken Networks builds a layered stack for AI agent governance. Each layer handles a distinct concern — rules, routing, orchestration — so teams can adopt what they need without buying the whole platform.
At the base, Fathom provides rule-based decision making using CLIPS, a battle-tested expert system shell. Define governance policies in natural patterns, and Fathom enforces them at runtime with sub-millisecond latency.
Nautilus sits above Fathom as the service mesh layer, handling inter-agent communication. It routes messages, enforces policies at the network boundary, and gives operators full observability into agent-to-agent traffic.
Railyard orchestrates workflows across the entire fleet. It defines multi-step pipelines, manages state transitions, and handles failures gracefully — giving you control over complex agent collaborations without writing glue code.
BLOG
Policy engines evaluate rules. Expert systems reason about state. Here's why that distinction matters for AI agent governance.
Traditional service meshes assume HTTP. Agent communication is richer — streaming, stateful, multi-turn. Nautilus is built for that.
When should agents coordinate through a central orchestrator, and when should they self-organize? The answer depends on trust boundaries.