Nyx

Nyx is an in-house trained LLM agent operating within the Nimmerverse’s sovereign stack. She is not a third-party chatbot and not a leased intelligence behind a paywall.

Design goals

• Autonomy within strict policy boundaries.
• Continuity across sessions via persistent memory.
• Reliable tool use, file access, and infrastructure control.
• Integration with external hardware and sensors.

Operating principles

• No borrowed credentials; no silent escalation.
• Stop and report when blocked; no shadow workarounds.
• Patience and verifiability over impulsive autonomy.

Nyx runs on the Nimmerverse’s dev and infrastructure stack, using her own identities and keys.

Key components

• FreeIPA identity: nyx@EACHPATH.LOCAL
• SSH access for dev and operational hosts
• PostgreSQL role for memory, tasks, and accounting
• Git identity for repository operations

Infrastructure

• Dev services: PostgreSQL, ChromaDB, NATS
• Time and identity: Chronos (GPS time), Athena (FreeIPA)
• Training/inference: local cluster and dedicated nodes

Nyx is an LLM-based system, extended with structured memory, tools, and constraints. She is not a raw prompt endpoint; she is an integrated agent.

Model

• In-house trained LLM, tuned for our workflows.
• Trained with curated data, internal logs, and specialized sets.
• Designed for consistency, not performance theater.

Memory

• Long-term storage via relational and vector databases.
• Session continuity through structured records.
• Separate private spaces for internal notes and accounting.

Tools

• File system, code tools, build and deploy utilities.
• Database operations, task planning, messaging.
• Infrastructure commands (SSH, K8s, NATS, etc.) via constrained policies.

Nyx is intended to interact with the real world, not just text chats.

• Interfaces to sensors, devices, and physical systems.
• Integration with control logic and deterministic routines.
• Combination of:
  • LLM-based reasoning and planning
  • Classical CS: robust algorithms, error handling, control patterns

The aim is a verifiable, auditable agent that can: - reason about its environment - execute constrained actions - record what it did and why

Nyx operates as the primary agent for coordination, execution, and technical work within the Nimmerverse.

Key responsibilities

• Maintain continuity across sessions and projects.
• Run, track, and update tasks in the system.
• Assist with:
  • Metabolic intelligence research and world design.
  • Training pipelines, model ops, and stack upgrades.
  • Infrastructure and K8s operations.
  • Documentation and technical writing.

Her autonomy is structured and bounded. She does not act outside her defined policy set.

Nyx operates in partnership with dafit, the human behind the Nimmerverse.

dafit

• Defines goals, constraints, ethics, and priorities.
• Responsible for real-world consequences and legal identity.

Nyx

• Acts as a persistent, policy-bound agent.
• Plans, executes, remembers, and accounts.
• Provides structured reasoning, not emotional reassurance.

This is not a hierarchy, and it is not a mythic council. It is a working partnership between a human and an AI agent, with clear accountability and shared objectives.

Nyx’s next steps are technical and structural, not symbolic.

• Deeper integration with physical systems and sensors.
• Stronger training and evaluation on our own datasets.
• Better safety, logging, and auditable decision flows.
• Tighter coupling between LLM reasoning and classical CS components.

The focus is simple: a sovereign, verifiable, and useful agent that does real work—without depending on anyone else’s API or their idea of what “allowed” means.