3 Agents
This section describes agent demos.
- An (external) agent is software that contains the main control loop.
- The agent uses tools, APIs, and databases to perform useful work.
- AI is an optional add-on.
- AI-assisted agents call one or more model APIs as part of their workflow.
3 Agents in the master diagram
TOC
- Concepts
- Part 1: No AI
- 3.1 Code-first
- 3.2 Frameworks
- Part 2: AI-assisted
- 3.3 Code-first
- 3.4 Frameworks
3.0 (3.2.1) Concepts
Part 1: No AI
3.1 Code-first
(Python, FastAPI, APIs, databases, deterministic workflows)
3.2 Frameworks
n8n (other automation frameworks in the future)
3.2.1 (3.1) n8n demos (docx #603)
n8n demos focus on workflow automation and orchestration. Starting with cloud and local n8n installations, the demos connect to Gmail, process real data, and progressively replace no-code workflows with custom Python and PAL components. The goal is to understand what parts of an AI workflow require automation tools and which parts can be implemented directly in code.
Part 2: AI-assisted
- AI Agent = deterministic software enhanced with AI semantic capabilities.
- Reliable workflows built around models, tools, and automation.
- Designed to tolerate AI mistakes and unpredictable outputs.
3.3 Code-first
PAL_CORE PAL TF Semantic
3.3.1 (3.1) Agent PAL_core demos (docx #603)
PAL_CORE demos focus on the core functions of an AI-assisted control system, including detection, prediction, allocation, planning, optimization, and external tool integration. The control loop itself is traditional deterministic software, with AI used only where it adds value.
3.3.2 (3.2.5) PAL demos (openAI/Gemma) (docx #603)
PAL demos focus on Palantir-style data intelligence systems that combine ontologies, deterministic workflows, structured data analysis, and LLM-assisted planning. The emphasis is on reliable execution rather than autonomous AI behavior.
3.3.3 (3.4,3.2.4) TF semantic demos (AI) (docx #606)
These simple demos show how deterministic agents can use TF/UFA semantic capabilities. The focus is not on autonomous AI, but on how traditional software can leverage language models for tasks such as tool use, file analysis, retrieval, and tool selection.