AI Infrastructure

MILTON utilizes an aggressive context-aware AI approach to render code structure into standard engineering specifications. This resides primarily in the backend MILTON.Infrastructure layer.

OpenAiCompatibleService

Implemented in OpenAiCompatibleService.cs, this class serves as an elastic HTTP bridge towards any OpenAI API contract-compliant endpoint. This enables MILTON setups across varying environments using:

  • Hosted OpenAI / Azure OpenAI endpoints.
  • Self-hosted localized inference instances using endpoints like vLLM, Ollama, or localized hardware inference endpoints.

Per-Preset Configurations

MILTON defines dynamic personas or “Presets” (e.g., Writer, Coder, Analyst). These overrides adjust the:

  • Targeting URL.
  • Model selection.
  • Execution temperatures.
  • Baseline system prompts.
  • Token context limits.

Privacy and Security

To comply with offline mandates and enterprise security bounds:

  1. API Keys are encrypted via AES-256 at rest.
  2. The infrastructure operates securely when paired with an air-gapped internal LLM runner, removing all necessity for outbound telemetric calls to standard CDN or public AI hosts.