---
title: "Technology | How Fitly AI works"
description: "Fitly is a ten-layer agent orchestrator with 50+ structured tools, full-context assembly, multi-model routing, and response guards. Not a chatbot bolted onto a tracker."
url: https://www.fitly.chat/technology
type: technology
---

# The engine behind your agent

*Most fitness AIs are a model and a prompt. Fitly is a ten-layer orchestrator: nine systems run before the model is allowed to speak, and a tenth runs after.*

## Why orchestration, not just call the model

A model is a smart, expensive, occasionally-wrong text engine. Everything that makes Fitly feel like an agent instead of a chatbot is engineered around it. The model is one component of the orchestrator, not the other way around.

## The ten layers that run on every message

1. **Auth, validation & usage limits** — every request is verified, schema-checked, and rate-limited before a token is spent.
2. **Image classification** — photos are sorted into eight types (meal, label, barcode, scale, equipment, body, screenshot, other) so each triggers the right pipeline.
3. **Parallel context assembly** — nine queries fan out at once: profile, macro targets, today's logs, recent workouts, active plans, history, workflow state, long-term memory, streaks.
4. **Workflow continuation detection** — decides whether a reply is a fresh request or a continuation, across nine multi-turn workflows.
5. **Dynamic system prompt builder** — the prompt is assembled per turn from 20+ conditional blocks.
6. **Dynamic tool scoping** — 50+ tools exist; the model only sees the few that fit the turn.
7. **Multi-model routing & self-escalation** — cheap turns run on a fast model, hard ones route to a stronger one, and a turn can escalate itself mid-flight.
8. **Agentic tool loop** — the model calls tools, reads results, and calls more, up to ten iterations, before it replies.
9. **Response guards** — write-confirmation, future-write, and partial-logging guards run before the reply ships.
10. **Persistence, memory & analytics** — the turn is stored, memory candidates extracted, streaks advanced. The layer that makes the next conversation sharper.

## What the orchestrator buys you

- **An agent, not a prompt** — it reasons, picks from 50+ structured tools, calls them in a loop, and checks the data before answering.
- **Full context, every turn** — nine queries assemble your present-tense state before the model speaks.
- **Provider-agnostic, multi-model** — runs on frontier models (Claude by default) behind an abstraction, routed per task.
- **Photo intelligence** — an eight-category classifier routes each image to the right pipeline.
- **Guards you can trust** — it can't lie about what it saved, log into future days, or drop a food.
- **Private by design** — encrypted health data, pre-signed URLs, and an audit log on every change.
- **Sharper every week** — after every turn, Fitly extracts what it learned about you into long-term memory.
- **Built to scale** — a stateless API, Redis-cached context, and queue-backed processing on AWS.

## An agent other agents can talk to

Fitly runs a Model Context Protocol (MCP) server. Your Claude, ChatGPT, Gemini, or any custom agent can connect and use the same 50+ tools the agent uses: reading your profile, macros, and history, and logging meals, workouts, weight, and water, with scoped, token-based access over Streamable HTTP. Your fitness data becomes a capability any agent can call, no app required. Available on the Fitly AI Pro plan.

## Go deeper

- The Orchestrator Inside Fitly (full deep dive): https://www.fitly.chat/blog/the-orchestrator-inside-fitly
- Features: https://www.fitly.chat/features

---

## Links

- [Live page](https://www.fitly.chat/technology)
- [Sign up](https://app.fitly.chat?mode=signup)
- [Marketing site](https://www.fitly.chat)
- [Short llms.txt](https://www.fitly.chat/llms.txt)