Reference

Glossary

Plain-language definitions of the terms we use across agentic AI and legacy modernization — so nothing on this site is a black box.

Back to home
Agentic AI / AI agent
Software that uses an LLM to plan and take actions through tools — reading, deciding, and acting on real systems — rather than just generating text.MCP connectors
AI DevOps / AIOps
Applying AI across the delivery pipeline: intelligent CI/CD, automated code review, and AI-driven monitoring, alerting, and incident triage.DevOps
Characterization tests
Tests that capture a legacy system’s current behavior so you can refactor confidently without changing what it does.DevOps
CI/CD
Continuous Integration / Continuous Delivery — the automated pipeline that builds, tests, and ships code changes safely and repeatably.DevOps
Evals
Repeatable tests that measure an AI system’s quality on real tasks, so changes can be judged objectively instead of by gut feel.DevOps
Fine-tuning
Further-training an open model on your own data and tasks so it performs better on your specific workload.Local LLMs
Guardrails
Constraints that keep agents safe: permissions, allowed actions, input/output validation, and limits on what a model can do.Services
Human-in-the-loop (HITL)
Workflows where a person reviews or approves consequential agent actions before they run.Services
Infrastructure as Code (IaC)
Managing servers and infrastructure through version-controlled code instead of manual setup, for reproducible environments.DevOps
Inference runtime
The server software (e.g. Ollama, vLLM) that actually runs a model and serves its responses.Local LLMs
Legacy migration
Modernizing decades-old codebases (COBOL, Java monoliths, PHP, .NET) into clean, testable, agent-ready services — without disrupting production.Process
LLM (Large Language Model)
The AI model (e.g. Claude, Gemma, Qwen) that powers language understanding and generation. It can run in the cloud or locally.Local LLMs
MCP (Model Context Protocol)
An open standard for connecting AI agents to real systems, exposing your apps, databases, and APIs as typed, permissioned tools any compliant agent can call.MCP connectors
MCP connector
A server that wraps an existing system in MCP, giving agents a safe, typed, permissioned interface to read from and act on it.MCP connectors
Observability
Tracing, logging, and metrics that make an AI system’s behavior visible, debuggable, and auditable in production.DevOps
On-prem / air-gapped
Running systems on infrastructure you control. Air-gapped means fully isolated from the public internet for maximum data security.Local LLMs
Open-weight / local model
A model whose weights you can download and run on your own hardware (Gemma, Qwen, Nemotron, Llama…), keeping data in-house.Local LLMs
Orchestration
Coordinating multiple agents, tools, and steps into a reliable workflow — controlling sequencing, retries, and hand-offs.Services
Quantization
Compressing a model to lower numerical precision so it runs faster and on smaller hardware, with minimal quality loss.Local LLMs
RAG (Retrieval-Augmented Generation)
Giving a model relevant, retrieved context (documents, data) at query time so answers are grounded and citable.Services
Rollback
The ability to quickly revert an AI-driven change or deployment when something goes wrong.DevOps
TDD (Test-Driven Development)
Writing tests before or alongside code so behavior is specified and protected — the safety net that makes automated and agentic changes trustworthy.DevOps