AI agent diagrams
A purpose-built diagram for LLM-powered systems: how a model orchestrates tools, retrieves context, remembers state, routes decisions, and hands off to a human.
What is an agent diagram?
An AI agent diagram models the control and data flow of an LLM application. Classic architecture nodes (services, databases) don’t capture the moving parts of an agent — the prompt, the tool calls, the retrieved context, the memory. Skeema gives each of these its own node type so you can design and reason about the agent’s behavior.
| Node | Role |
|---|---|
LLM | The model call — system prompt, temperature, model id. |
Tool | A function the agent can invoke (search, code-exec, API call). |
Retriever | Fetches context (vector search, lookup) to ground the model. |
Memory | Short- or long-term state carried across turns. |
Router | Branches the flow based on the model’s decision or a classifier. |
Human-in-Loop | A checkpoint where a person approves or edits before continuing. |
Simulating an agent
Agent diagrams support step-by-step playback. Press play and Skeema walks the graph one node at a time, marking each as active, completed, failed, or retrying, and highlighting the active edge. Control speed (0.5× / 1× / 2×) or step manually. This makes the agent’s control flow — especially routing and retries — concrete and reviewable.
⌘/Ctrl + Shift + L to re-layout after edits.When to use one
- •Designing a RAG or tool-using agent before writing the orchestration code.
- •Communicating an agent’s decision flow to teammates or reviewers.
- •Deciding where a human checkpoint belongs in an automated pipeline.
- ✓Agent diagrams model LLM control flow: model, tools, retrievers, memory, routers, human steps.
- ✓Step-by-step playback shows how the agent actually traverses the graph.
- ✓Each node is configurable — model, temperature, tool signature, retriever source.