Skip to main content

Documentation Index

Fetch the complete documentation index at: https://reliatrack.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

Agent orchestration is the process of coordinating everything that happens between you typing a request and your AI agent delivering a final answer. In ApexSpriteAI, that coordination is handled by the Claude Code CLI, which sits at the center of your workflow: it takes your input, sends it to the AI model, interprets what the model wants to do next, executes any tools the model requests, and then feeds those results back until the task is complete. Understanding this cycle helps you build more reliable, capable agents.

How the orchestration cycle works

Every interaction in ApexSpriteAI follows the same repeating loop. The Claude Code CLI drives each step, acting as the bridge between you, the AI model, and the tools available in your environment.
1

You send a request

You type a prompt in your terminal — for example, claude "Add a new endpoint to my Express app". The Claude Code CLI packages your message along with descriptions of all available MCP tools and sends the combined payload to the AI model.
2

The model runs inference

The AI model (running in LM Studio on your GPU hardware) receives the prompt and generates a response. This response is either a direct answer or a decision to use a tool before answering.
3

The CLI checks for a tool call

The Claude Code CLI parses the model’s response. If the model returned a plain text answer, the cycle ends and your result is displayed. If the model returned a tool call, the CLI moves to the next step.
4

The CLI executes the tool locally

Tool execution always happens on your local machine, not on the remote model server. The CLI runs the requested tool — for example, reading a file, running a shell command, or calling an API — and captures the output.
5

The result is fed back to the model

The CLI sends the tool result back to the model as part of the ongoing conversation context. The model now has the information it requested and can continue reasoning.
6

The model generates the final response

With the tool result in context, the model produces its final answer. The CLI displays it in your terminal. If the model needs additional tool calls, the cycle repeats from step 3.

The Claude Code CLI as orchestrator

The Claude Code CLI is not just a chat interface — it is the orchestrator that keeps every part of your agent stack in sync. It manages three responsibilities simultaneously:

Context management

The CLI maintains the full conversation history, including tool results, so the model always has the context it needs to make good decisions.

Model communication

The CLI formats requests in the Anthropic messages API format and routes them to whichever inference backend you have configured, whether that is a remote API or your local LM Studio server.

Tool execution

When the model decides to use a tool, the CLI executes it locally on your machine. The model only describes what it wants to do; the CLI actually does it.

Loop control

The CLI decides when to keep looping (more tool calls needed) and when to stop (final answer reached), preventing runaway execution.

MCP: the tool integration standard

ApexSpriteAI uses the Model Context Protocol (MCP) as the standard for defining and connecting tools to your agent. MCP gives the AI model a consistent way to declare which tools are available, what parameters they accept, and what they return. This standardization means you can add new capabilities to your agent — file access, web search, database queries, and more — without changing how the orchestration loop works.
Because tool execution always happens on your local machine via the Claude Code CLI, the location of your AI model does not limit what tools your agent can use. A model running on a remote GPU server has full access to your local file system, shell, and any MCP tools you have installed.
To see which MCP tools your agent currently has access to, run claude mcp list in your terminal. Each listed tool becomes part of the context the model receives on every request.