AI Orchestration
Design, run, test, and monitor multi-agent AI pipelines from one platform — Agent Architect, Prompt Lab, and full observability, all wired into every workflow in TotalApp.
Overview
AI Orchestration is where TotalApp's AI agents get designed, tested, and monitored. Rather than treating each agent as a one-off integration, orchestration gives you a visual canvas to build multi-step pipelines, a sandbox to test prompts before they go live, and a single dashboard that logs every model call made anywhere in the platform.
An agent built here isn't limited to a chat window — once designed, it can be exported as a reusable node and dropped directly into the Workflow Editor, so the same reasoning or automation step can run inside any business process, not just a standalone assistant.
Where to configure it
Agent Architect, Prompt Lab, and Observability are reached from the AI Orchestration workspace inside TotalApp. See Agentic AI Settings for the model configuration that backs every agent, Local Language Models and Cloud LLM Models for the model catalog an agent can pick from, and AI Governance for how permissions and audit logging apply to everything built here.
Agent Architect
Pick an LLM, define prompt chains, wire in tools, and connect agents to any workflow — all in a visual canvas. Agent Architect is where a pipeline goes from an idea to something a workflow can call.
- Pick from local or cloud LLMs per agent — mix models in one pipeline depending on cost and capability.
- Build multi-step prompt pipelines visually with drag-and-drop nodes instead of hand-written orchestration code.
- Export a finished agent as a reusable node directly in the Workflow Editor.
- Choose from Standard, Reasoning, Autonomous, Multimodal, and Domain-specific agent types depending on what the task needs.
Prompt Lab
Write, run, and compare prompts across any configured model in a live sandbox before wiring them into an agent — no copy-pasting a prompt between tools to see how it behaves on a different model.
- Run a prompt against any active model — local or cloud — in one click.
- Compare outputs side-by-side across models with diff highlighting, to see which one actually performs better on your data.
- Save a prompt template and inject it directly into an Agent Architect node once it's proven out.
- Token count, latency, and cost are shown per run, so a model choice is never a guess.
Observability & Logs
Every model call across every TotalApp screen is logged automatically — tokens consumed, tools called, latency, errors — in one dashboard, whether the call came from an agent built in Agent Architect or a built-in AI feature elsewhere in the platform.
Per-event log
Token breakdown, duration, tool calls, and error trace for every individual model call, not just an aggregate count.
Cross-application analytics
Token usage broken down by source app, model, and user — every AI screen logs here automatically, with nothing to wire up manually.
Exportable logs
Export activity as CSV for cost reporting and compliance audits, the same data an AI Governance review would draw on.
Every AI Layer, One Platform
Design, run, test, monitor, and extend AI agents without leaving TotalApp — orchestration covers the full lifecycle of an agent, from first prompt to production workflow node.
| Capability | What it does |
|---|---|
| Agent Architect | Design multi-step AI agents visually — pick LLM, wire prompt chains, attach tools, export as workflow nodes. |
| Model Center | Manage every local and cloud LLM from one screen; switch the active model for any agent with a click. |
| Prompt Lab | Test and refine prompts in a live sandbox, comparing outputs across models before production use. |
| Knowledge Base | Upload documents, convert to vector embeddings, and expose them to any agent node as RAG context. |
| Agent Persona Editor | Define an agent's character, capabilities, and constraints with a detailed system prompt. |
| Tool / Function Registry | Define JSON Schema-based tools — database queries, API calls, file I/O, email — and wire them into any agent. |
| Observability & Logs | Every AI call across every screen logged — tokens, latency, tool calls, errors — in one dashboard. |
| Pre-built Agent Nodes | Drop Standard, Reasoning, Autonomous, Multimodal, and Domain-specific agent nodes into any workflow. |
From Prompt to Production
A pipeline never has to be rebuilt to move from experiment to production — the same node built in Agent Architect is what a workflow calls, and the same Observability dashboard tracks it once it's live.