TallMountain – Stoic Virtue Ethics for an LLM Agent (github.com)

TallMountain-Raku is an open-source Raku framework that wraps an LLM with a formally specified, Stoic-inspired machine ethics system. Drawing on Lawrence C. Becker’s A New Stoicism, the project translates virtue ethics into a computable normative calculus that filters all LLM outputs. Its three-tier hierarchy—virtue propositions (non-negotiable moral requirements), norms (rules and duties derived from virtues), and externals (preferred but non-binding outcomes)—is enforced with deontic operators (OUGHT, OUGHT_NOT, PERMITTED) and a lexical ordering that guarantees virtues override norms and externals. Significance: this gives agent builders a transparent, auditable way to encode ethical priorities and resolve conflicts in context-sensitive ways, helping mitigate harms like prompt injection, unsafe advice, or user vulnerability. Technically, TallMountain runs a cognitive cycle of reactive (threat scanning), deliberative (planning), and normative (ethical evaluation) stages. Key components live under lib/Normative/ and lib/Cycle/, with configurable threat scanners, REPL/web UI and REST endpoints, and JSON-based normative propositions (config/system-endeavours.json). The repo (github.com/seamus-brady/tallmountain-raku) includes Docker and local-run instructions; thresholds and rules are editable for testing and deployment. Active development invites contributions for richer normative examples, conflict-resolution tests, and UX improvements. Licensed MIT, it’s positioned as a practical research and engineering platform for safer, ethics-first LLM agents.